{
 "benches": [
  {
   "bench": "reading-text",
   "title": "Reading · Turkish",
   "url": "https://zekiworks.github.io/analysis/decision-models/",
   "updated": "2026-11-16T20:50:26Z",
   "score": {
    "zero": "chance: 1 of 5 options right on every question (accuracy 20%)",
    "hundred": "every one of the 2,198 Turkish reading and grammar questions right (accuracy 100%)"
   },
   "models": [
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Carbon",
     "tag": "high · chat",
     "model_key": "gemini-4-carbon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 98.06642402183805,
     "raw": {
      "metric": "accuracy",
      "value": 0.9845313921747043,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0042720655141021985,
     "improvement_points": null,
     "runs": [
      "text:1108"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.2",
     "tag": "max · chat",
     "model_key": "claude-fable-5.2",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.6683348498635,
     "raw": {
      "metric": "accuracy",
      "value": 0.9813466787989081,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0753002729754179,
     "improvement_points": null,
     "runs": [
      "text:980"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 3.8 Flash",
     "tag": "high · chat",
     "model_key": "gemini-3.8-flash",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.61146496815287,
     "raw": {
      "metric": "accuracy",
      "value": 0.9808917197452229,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.010841674249322482,
     "improvement_points": null,
     "runs": [
      "text:140",
      "text:123",
      "text:188"
     ],
     "notes": "Gemini API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Hydrogen",
     "tag": "high · chat",
     "model_key": "gemini-4-hydrogen",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.61146496815287,
     "raw": {
      "metric": "accuracy",
      "value": 0.9808917197452229,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.011883530482249363,
     "improvement_points": null,
     "runs": [
      "text:1114"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Astra",
     "tag": "high · chat",
     "model_key": "gpt-6-astra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.44085532302094,
     "raw": {
      "metric": "accuracy",
      "value": 0.9795268425841674,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0065000000000037935,
     "improvement_points": null,
     "runs": [
      "text:277",
      "text:610",
      "text:48",
      "text:191"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Sol",
     "tag": "high · chat",
     "model_key": "gpt-6.1-sol",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.21337579617834,
     "raw": {
      "metric": "accuracy",
      "value": 0.9777070063694268,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.009818016378528189,
     "improvement_points": null,
     "runs": [
      "text:276",
      "text:124",
      "text:190"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Argon",
     "tag": "low · chat",
     "model_key": "gemini-4-argon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.4504094631483,
     "raw": {
      "metric": "accuracy",
      "value": 0.9636032757051866,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.02185623293903984,
     "improvement_points": null,
     "runs": [
      "text:557"
     ],
     "notes": "Gemini API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.1",
     "tag": "high · chat",
     "model_key": "claude-fable-5.1",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.16606005459506,
     "raw": {
      "metric": "accuracy",
      "value": 0.9613284804367607,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.11954504094631425,
     "improvement_points": null,
     "runs": [
      "text:310"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Opus 5.5",
     "tag": "high · chat",
     "model_key": "claude-opus-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 93.74431301182892,
     "raw": {
      "metric": "accuracy",
      "value": 0.9499545040946314,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.14620109190173014,
     "improvement_points": null,
     "runs": [
      "text:278",
      "text:47",
      "text:192"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Terra",
     "tag": "high · chat",
     "model_key": "gpt-6.1-terra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 92.8343949044586,
     "raw": {
      "metric": "accuracy",
      "value": 0.9426751592356688,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.02824840764332354,
     "improvement_points": null,
     "runs": [
      "text:275",
      "text:274"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Grok 4.7",
     "tag": "high · chat",
     "model_key": "grok-4.7",
     "maker": "xAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 90.90081892629665,
     "raw": {
      "metric": "accuracy",
      "value": 0.9272065514103731,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.11871246587807233,
     "improvement_points": null,
     "runs": [
      "text:311"
     ],
     "notes": "xAI Responses API, SuperGrok subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · first release",
     "model_key": "deepseek-v4.1-flash-first-release",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": 552.0,
     "params_active": 16.0,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 90.04777070063693,
     "raw": {
      "metric": "accuracy",
      "value": 0.9203821656050956,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.07960198328526213,
     "improvement_points": null,
     "runs": [
      "text:121",
      "text:120"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; first-release weights (revision fb2764a), not the 1017 update"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · 1017",
     "model_key": "deepseek-v4.1-flash-1017",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 89.70655141037305,
     "raw": {
      "metric": "accuracy",
      "value": 0.9176524112829846,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.06880119663902093,
     "improvement_points": null,
     "runs": [
      "text:585",
      "text:558"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; 1017 weights (revision 1bae0e1)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen4-Max",
     "tag": "high · chat",
     "model_key": "qwen4-max",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 89.64968152866243,
     "raw": {
      "metric": "accuracy",
      "value": 0.9171974522292994,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0784672206832799,
     "improvement_points": null,
     "runs": [
      "text:1336"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint (the maker's), fallbacks off, reasoning effort high, output cap 32,768 tokens, reasoning included; the answer schema"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Muse Ember",
     "tag": "max · chat",
     "model_key": "muse-ember",
     "maker": "Meta",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.02911737943586,
     "raw": {
      "metric": "accuracy",
      "value": 0.8562329390354868,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0616418604651154,
     "improvement_points": null,
     "runs": [
      "text:939"
     ],
     "notes": "OpenRouter chat API, Meta's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning max"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Sonnet 5.5",
     "tag": "high · chat",
     "model_key": "claude-sonnet-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 81.28980891719745,
     "raw": {
      "metric": "accuracy",
      "value": 0.8503184713375797,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.15961328480436646,
     "improvement_points": null,
     "runs": [
      "text:55",
      "text:193",
      "text:54"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Kimi K3.1 Pro",
     "tag": "low · chat",
     "model_key": "kimi-k3.1-pro",
     "maker": "Moonshot",
     "kind": "chat",
     "params_total": 2800.0,
     "params_active": 104.0,
     "quantization": null,
     "score_0_100": 78.44631483166515,
     "raw": {
      "metric": "accuracy",
      "value": 0.8275705186533212,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1256551410373169,
     "improvement_points": null,
     "runs": [
      "text:320",
      "text:315"
     ],
     "notes": "OpenRouter chat API, Moonshot AI's endpoint (mxfp4), fallbacks off"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.5-397B-A17B",
     "tag": "high · chat",
     "model_key": "qwen-3.5-397b-a17b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 397.0,
     "params_active": 17.0,
     "quantization": null,
     "score_0_100": 78.3894449499545,
     "raw": {
      "metric": "accuracy",
      "value": 0.827115559599636,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.062252047315757925,
     "improvement_points": null,
     "runs": [
      "text:870"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included"
    },
    {
     "repo": "Poindexter-Labs/Copernicus-31B",
     "commit": "be71525",
     "name": "Copernicus-31B",
     "tag": "",
     "model_key": "copernicus-31b",
     "maker": "Poindexter Labs",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 77.25204731574159,
     "raw": {
      "metric": "accuracy",
      "value": 0.8180163785259327,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.09057793964924428,
     "improvement_points": null,
     "runs": [
      "text:1165"
     ],
     "notes": "Copernicus server (the card's copernicus_server.py, Transformers), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen1.5",
     "commit": "8e56aa1b236a6dd1064dcfd22ba1e5e360e4c140",
     "name": "KnowLine-31B-Gen1.5",
     "tag": "",
     "model_key": "knowline-31b-gen1.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 77.25204731574159,
     "raw": {
      "metric": "accuracy",
      "value": 0.8180163785259327,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03390679080566063,
     "improvement_points": null,
     "runs": [
      "text:957"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen2",
     "commit": "fd99a9e",
     "name": "KnowLine-31B-Gen2",
     "tag": "",
     "model_key": "knowline-31b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 76.51273885350318,
     "raw": {
      "metric": "accuracy",
      "value": 0.8121019108280255,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.05637409125764909,
     "improvement_points": null,
     "runs": [
      "text:1314"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "turion-ai/turion-jev-31b",
     "commit": "866c0fc6e040cba23fcc72113a80e5c99d5ebab1",
     "name": "Turion Jev 31B",
     "tag": "",
     "model_key": "turion-jev-31b",
     "maker": "Turion.AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 at load",
     "score_0_100": 75.14786169244766,
     "raw": {
      "metric": "accuracy",
      "value": 0.8011828935395814,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1111902520473159,
     "improvement_points": null,
     "runs": [
      "text:1273"
     ],
     "notes": "Turion Jev server (turion_jev.serve on vLLM, the maker's release code), self-hosted, FP8 at load"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Inkling-Small",
     "tag": "high · chat · Jul 2026",
     "model_key": "inkling-small",
     "maker": "Thinking Machines",
     "kind": "chat",
     "params_total": 265.96,
     "params_active": 12.0,
     "quantization": null,
     "score_0_100": 74.97725204731573,
     "raw": {
      "metric": "accuracy",
      "value": 0.7998180163785259,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.07924476797088931,
     "improvement_points": null,
     "runs": [
      "text:866"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included; no response format (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3-Flash",
     "tag": "reasoning on · chat",
     "model_key": "glm-5.3-flash",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 320.0,
     "params_active": 18.0,
     "quantization": "EXL3 4-bit experts, FP8 dense layers and KV cache",
     "score_0_100": 73.49863512283893,
     "raw": {
      "metric": "accuracy",
      "value": 0.7879890809827116,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.04548309178743938,
     "improvement_points": null,
     "runs": [
      "text:160",
      "text:159"
     ],
     "notes": "TensorFold, self-hosted, EXL3 4-bit experts"
    },
    {
     "repo": "decisionmodels/REAX-1.3",
     "commit": "fa4bf7e",
     "name": "REAX 1.3",
     "tag": "",
     "model_key": "reax-1.3",
     "maker": "REAX",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 72.41810737033666,
     "raw": {
      "metric": "accuracy",
      "value": 0.7793448589626933,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.04139613055428547,
     "improvement_points": null,
     "runs": [
      "text:827"
     ],
     "notes": "REAX Engine server (reax13 mode, the maker's release code), self-hosted, FP8"
    },
    {
     "repo": "MachinePulseAI/karpo-decision-31B",
     "commit": "d69cdd5",
     "name": "Karpo Decision 31B",
     "tag": "",
     "model_key": "karpo-decision-31b",
     "maker": "MachinePulse",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 72.24749772520472,
     "raw": {
      "metric": "accuracy",
      "value": 0.7779799818016379,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.02332295486319941,
     "improvement_points": null,
     "runs": [
      "text:1156"
     ],
     "notes": "karpo-decision server (the System One server of the card's karpo-decision package, machinepulse-ai/karpo-decision at ac40d39), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini Decision 1",
     "tag": "20261111",
     "model_key": "gemini-decision-1",
     "maker": "Google",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 71.90627843494084,
     "raw": {
      "metric": "accuracy",
      "value": 0.7752502274795269,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.02027338817569684,
     "improvement_points": null,
     "runs": [
      "text:1021"
     ],
     "notes": "OpenRouter decisions API, Google AI Studio endpoint"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1-ultra",
     "tag": "",
     "model_key": "liquid-d1-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 71.33757961783441,
     "raw": {
      "metric": "accuracy",
      "value": 0.7707006369426752,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03629871483620005,
     "improvement_points": null,
     "runs": [
      "text:199"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": "logitshq/Logits-Sigma-1-Duo",
     "commit": "a0c1b6e",
     "name": "Logits Sigma-1 Duo",
     "tag": "",
     "model_key": "logits-sigma-1-duo",
     "maker": "Logits",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 68.7784349408553,
     "raw": {
      "metric": "accuracy",
      "value": 0.7502274795268425,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.05678135723610731,
     "improvement_points": null,
     "runs": [
      "text:1221"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "Qwen/Qwen3.8-27B",
     "commit": "1d4bf0f",
     "name": "Qwen3.8-27B",
     "tag": "low · chat",
     "model_key": "qwen3.8-27b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 27.78,
     "params_active": 27.78,
     "quantization": "BF16",
     "score_0_100": 68.32347588717016,
     "raw": {
      "metric": "accuracy",
      "value": 0.7465878070973613,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.23697412193898643,
     "improvement_points": null,
     "runs": [
      "text:318",
      "text:50",
      "text:53",
      "text:83"
     ],
     "notes": "vLLM, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3",
     "tag": "low · chat",
     "model_key": "glm-5.3",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 753.33,
     "params_active": 753.33,
     "quantization": null,
     "score_0_100": 68.1528662420382,
     "raw": {
      "metric": "accuracy",
      "value": 0.7452229299363057,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.025787079162879617,
     "improvement_points": null,
     "runs": [
      "text:314"
     ],
     "notes": "OpenRouter chat API, Fireworks endpoint, fallbacks off"
    },
    {
     "repo": "sthanika-ai/Sieve-31B-v3",
     "commit": "8fcee858f7f20a50bfa3c8f984c22adb3cbb46a2",
     "name": "Sieve 31B v3",
     "tag": "",
     "model_key": "sieve-31b-v3",
     "maker": "Sthānika AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 67.69790718835306,
     "raw": {
      "metric": "accuracy",
      "value": 0.7415832575068244,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.028719244767970685,
     "improvement_points": null,
     "runs": [
      "text:1241"
     ],
     "notes": "Sieve server (sieve-serve, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "google/gemma-4-31B-it",
     "commit": "842da37",
     "name": "Gemma 4 31B",
     "tag": "reasoning off · chat",
     "model_key": "gemma-4-31b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 weights and KV cache",
     "score_0_100": 66.38989990900818,
     "raw": {
      "metric": "accuracy",
      "value": 0.7311191992720655,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.24156780736968866,
     "improvement_points": null,
     "runs": [
      "text:119",
      "text:189"
     ],
     "notes": "vLLM, self-hosted, FP8"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiMo-V2.6-Flash",
     "tag": "high · chat",
     "model_key": "mimo-v2.6-flash",
     "maker": "Xiaomi",
     "kind": "chat",
     "params_total": 309.0,
     "params_active": 15.0,
     "quantization": null,
     "score_0_100": 65.13876251137398,
     "raw": {
      "metric": "accuracy",
      "value": 0.7211101000909919,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.07140581068416227,
     "improvement_points": null,
     "runs": [
      "text:877"
     ],
     "notes": "OpenRouter chat API, Xiaomi's own endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included"
    },
    {
     "repo": "fastino/GLiDE-27B",
     "commit": "4d32743",
     "name": "GLiDE 27B",
     "tag": "",
     "model_key": "glide-27b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 63.887625113739766,
     "raw": {
      "metric": "accuracy",
      "value": 0.7111010009099181,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.023139052123107375,
     "improvement_points": null,
     "runs": [
      "text:899"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.2",
     "tag": "API · with web",
     "model_key": "perplexity-decider-27b-v1.2-with-web",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 60.93039126478617,
     "raw": {
      "metric": "accuracy",
      "value": 0.6874431301182894,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03589856878495877,
     "improvement_points": null,
     "runs": [
      "text:640"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": "fastino/GLiDE",
     "commit": "a7e4589",
     "name": "GLiDE",
     "tag": "v1.2",
     "model_key": "fastino-glide-v1.2",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 60.58917197452228,
     "raw": {
      "metric": "accuracy",
      "value": 0.6847133757961783,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.029152681148497294,
     "improvement_points": null,
     "runs": [
      "text:248"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.3",
     "commit": "9c2ebb2",
     "name": "Perplexity Decider 27B v1.3",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 60.532302092811655,
     "raw": {
      "metric": "accuracy",
      "value": 0.6842584167424932,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.049660688083001704,
     "improvement_points": null,
     "runs": [
      "text:1134"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "low · chat",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 60.361692447679715,
     "raw": {
      "metric": "accuracy",
      "value": 0.6828935395814377,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.2565468607825413,
     "improvement_points": null,
     "runs": [
      "text:133"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2-ultra",
     "tag": "",
     "model_key": "liquid-d1.2-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 60.13421292083713,
     "raw": {
      "metric": "accuracy",
      "value": 0.681073703366697,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.034231525406292845,
     "improvement_points": null,
     "runs": [
      "text:1347"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLiDE",
     "tag": "API",
     "model_key": "fastino-glide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 59.84986351228388,
     "raw": {
      "metric": "accuracy",
      "value": 0.6787989080982711,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.04508679278866653,
     "improvement_points": null,
     "runs": [
      "text:129"
     ],
     "notes": "Fastino System One API"
    },
    {
     "repo": "vllm-sr/d3",
     "commit": "20f37b9",
     "name": "d3",
     "tag": "",
     "model_key": "vllm-sr-d3",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 59.7361237488626,
     "raw": {
      "metric": "accuracy",
      "value": 0.6778889899909009,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0260618026254459,
     "improvement_points": null,
     "runs": [
      "text:259"
     ],
     "notes": "vLLM System One endpoint (nightly), self-hosted, BF16"
    },
    {
     "repo": "vllm-sr/d3-ultra",
     "commit": "52ba9e7",
     "name": "d3-ultra",
     "tag": "",
     "model_key": "vllm-sr-d3-ultra",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 59.7361237488626,
     "raw": {
      "metric": "accuracy",
      "value": 0.6778889899909009,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.12470444406965506,
     "improvement_points": null,
     "runs": [
      "text:269"
     ],
     "notes": "d3 package server (transformers, the release code), self-hosted, BF16"
    },
    {
     "repo": "nace-ai/drex-max-v1",
     "commit": "cac65e1",
     "name": "Drex Max",
     "tag": "",
     "model_key": "drex-max-v1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 59.67925386715195,
     "raw": {
      "metric": "accuracy",
      "value": 0.6774340309372157,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.06325311335192765,
     "improvement_points": null,
     "runs": [
      "text:322"
     ],
     "notes": "Drex Max server (Nace's loader, drex_max), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex Max v1.1",
     "tag": "",
     "model_key": "drex-max-1.1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 59.224294813466784,
     "raw": {
      "metric": "accuracy",
      "value": 0.6737943585077343,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.02920837835252908,
     "improvement_points": null,
     "runs": [
      "text:326"
     ],
     "notes": "Nace.AI System One API"
    },
    {
     "repo": "baselabs/mammoth-1",
     "commit": "61cec44",
     "name": "Mammoth-1",
     "tag": "",
     "model_key": "mammoth-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 58.88307552320291,
     "raw": {
      "metric": "accuracy",
      "value": 0.6710646041856233,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.05571809872611476,
     "improvement_points": null,
     "runs": [
      "text:374"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.2",
     "commit": "225c913",
     "name": "Perplexity Decider 27B v1.2",
     "tag": "DGX Spark",
     "model_key": "perplexity-decider-27b-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 58.82620564149226,
     "raw": {
      "metric": "accuracy",
      "value": 0.6706096451319381,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0443262835543908,
     "improvement_points": null,
     "runs": [
      "text:217",
      "text:633"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 58.598726114649665,
     "raw": {
      "metric": "accuracy",
      "value": 0.6687898089171974,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03844291588122355,
     "improvement_points": null,
     "runs": [
      "text:145"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiniMax-M3.1-Flash",
     "tag": "high · chat",
     "model_key": "minimax-m3.1-flash",
     "maker": "MiniMax",
     "kind": "chat",
     "params_total": 428.0,
     "params_active": 23.0,
     "quantization": null,
     "score_0_100": 58.08689717925386,
     "raw": {
      "metric": "accuracy",
      "value": 0.664695177434031,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.2113114754098455,
     "improvement_points": null,
     "runs": [
      "text:862"
     ],
     "notes": "OpenRouter chat API, MiniMax's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning included; JSON mode without the answer schema (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap Preview",
     "tag": "",
     "model_key": "subq-snap-preview",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 56.551410373066425,
     "raw": {
      "metric": "accuracy",
      "value": 0.6524112829845314,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.048948969200484,
     "improvement_points": null,
     "runs": [
      "text:177"
     ],
     "notes": "SubQ Decisions API (public preview)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap 1.1",
     "tag": "",
     "model_key": "subq-snap-1.1",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 54.84531392174704,
     "raw": {
      "metric": "accuracy",
      "value": 0.6387625113739763,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.11150360240298471,
     "improvement_points": null,
     "runs": [
      "text:346"
     ],
     "notes": "SubQ Decisions API (public preview)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "Decisions API",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 54.50409463148317,
     "raw": {
      "metric": "accuracy",
      "value": 0.6360327570518653,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.023755697356427245,
     "improvement_points": null,
     "runs": [
      "text:143"
     ],
     "notes": "OpenAI Decisions API (public beta)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.8 Flash",
     "tag": "no reasoning · chat",
     "model_key": "qwen3.8-flash",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 53.821656050955404,
     "raw": {
      "metric": "accuracy",
      "value": 0.6305732484076433,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.2872884440400548,
     "improvement_points": null,
     "runs": [
      "text:985"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint, fallbacks off, reasoning off"
    },
    {
     "repo": "PelaAI/KnowLine-12B-Gen1",
     "commit": "f26fbc2",
     "name": "KnowLine-12B-Gen1",
     "tag": "",
     "model_key": "knowline-12b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 53.08234758871701,
     "raw": {
      "metric": "accuracy",
      "value": 0.6246587807097361,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.06193311159534239,
     "improvement_points": null,
     "runs": [
      "text:1309"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2",
     "tag": "",
     "model_key": "liquid-d1.2",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 52.17242948134668,
     "raw": {
      "metric": "accuracy",
      "value": 0.6173794358507735,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03341676714248475,
     "improvement_points": null,
     "runs": [
      "text:1346"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.1",
     "tag": "",
     "model_key": "liquid-d1.1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 51.88808007279344,
     "raw": {
      "metric": "accuracy",
      "value": 0.6151046405823476,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.019863492797650786,
     "improvement_points": null,
     "runs": [
      "text:254"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "celeris-1-decision",
     "tag": "",
     "model_key": "celeris-1-decision",
     "maker": "Celeris",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 50.58007279344858,
     "raw": {
      "metric": "accuracy",
      "value": 0.6046405823475887,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.04257051865332119,
     "improvement_points": null,
     "runs": [
      "text:150"
     ],
     "notes": "Celeris System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 50.466333030027286,
     "raw": {
      "metric": "accuracy",
      "value": 0.6037306642402184,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.049916517793603135,
     "improvement_points": null,
     "runs": [
      "text:44"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": "OmniJev/OneJev-27B",
     "commit": "a5e5f55",
     "name": "OneJev-27B",
     "tag": "",
     "model_key": "onejev-27b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 49.499545040946316,
     "raw": {
      "metric": "accuracy",
      "value": 0.5959963603275705,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.05534345950864414,
     "improvement_points": null,
     "runs": [
      "text:264"
     ],
     "notes": "OneJev server (qev, OmniJev's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.14",
     "tag": "",
     "model_key": "jev-1.14.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 48.98771610555049,
     "raw": {
      "metric": "accuracy",
      "value": 0.591901728844404,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.025313921747042376,
     "improvement_points": null,
     "runs": [
      "text:194"
     ],
     "notes": "TypeSafe System One API"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning on · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 48.64649681528662,
     "raw": {
      "metric": "accuracy",
      "value": 0.589171974522293,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1149"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking on; output cap 32,768"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Mercury Decide",
     "tag": "",
     "model_key": "inception-mercury-decide",
     "maker": "Inception",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 47.85031847133757,
     "raw": {
      "metric": "accuracy",
      "value": 0.5828025477707006,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.15503569618513982,
     "improvement_points": null,
     "runs": [
      "text:204"
     ],
     "notes": "OpenRouter System One API"
    },
    {
     "repo": "google/gemma-4-E4B-it",
     "commit": "4ecc2ba",
     "name": "Gemma 4 E4B",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-e4b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 8.0,
     "params_active": 8.0,
     "quantization": "BF16",
     "score_0_100": 46.94040036396724,
     "raw": {
      "metric": "accuracy",
      "value": 0.575523202911738,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1040"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E4B-it@4ecc2ba), one question per request, plain answer line CEVAP: X (no answer schema); sampling temperature 0.0; output cap 8,192"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.13",
     "tag": "",
     "model_key": "jev-1.13.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 46.88353048225659,
     "raw": {
      "metric": "accuracy",
      "value": 0.5750682438580528,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.051401273885350714,
     "improvement_points": null,
     "runs": [
      "text:49"
     ],
     "notes": "TypeSafe System One API"
    },
    {
     "repo": "Cloudflare/clef",
     "commit": "2f3de3d",
     "name": "Clef",
     "tag": "",
     "model_key": "clef",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 27.48,
     "params_active": 27.48,
     "quantization": "BF16",
     "score_0_100": 46.599181073703356,
     "raw": {
      "metric": "accuracy",
      "value": 0.5727934485896269,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1069694722474975,
     "improvement_points": null,
     "runs": [
      "text:125"
     ],
     "notes": "Clef server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Haiku 5.5",
     "tag": "max · chat",
     "model_key": "claude-haiku-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 46.31483166515014,
     "raw": {
      "metric": "accuracy",
      "value": 0.5705186533212011,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.17499292786421325,
     "improvement_points": null,
     "runs": [
      "text:948"
     ],
     "notes": "OpenRouter chat API, Anthropic's own endpoint, fallbacks off, output cap 128,000 tokens, reasoning max"
    },
    {
     "repo": "tattrongvu/Jeffan-9B-1211",
     "commit": "862d1a7",
     "name": "Jeffan-9B-1211",
     "tag": "",
     "model_key": "jeffan-9b-1211",
     "maker": "tattrongvu",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 46.257961783439484,
     "raw": {
      "metric": "accuracy",
      "value": 0.5700636942675159,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.04648128028635688,
     "improvement_points": null,
     "runs": [
      "text:1192"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1",
     "tag": "",
     "model_key": "liquid-d1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 46.257961783439484,
     "raw": {
      "metric": "accuracy",
      "value": 0.5700636942675159,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.10208650451299976,
     "improvement_points": null,
     "runs": [
      "text:52"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex 2.0",
     "tag": "",
     "model_key": "drex-2.0",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 35.11,
     "params_active": 3.0,
     "quantization": null,
     "score_0_100": 44.665605095541395,
     "raw": {
      "metric": "accuracy",
      "value": 0.5573248407643312,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.042306960873521385,
     "improvement_points": null,
     "runs": [
      "text:334"
     ],
     "notes": "Nace.AI System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Cygnet",
     "tag": "",
     "model_key": "cygnet",
     "maker": "Blockbrain Labs",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 44.60873521383075,
     "raw": {
      "metric": "accuracy",
      "value": 0.556869881710646,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.137566046929174,
     "improvement_points": null,
     "runs": [
      "text:127"
     ],
     "notes": "Cygnet decision server on vLLM, self-hosted"
    },
    {
     "repo": "ZefanCai/Open-Jev-27B-v1.1",
     "commit": "28cf730",
     "name": "Open-Jev-27B-v1.1",
     "tag": "",
     "model_key": "open-jev-27b-v1.1",
     "maker": "Zefan Cai",
     "kind": "decision",
     "params_total": 27.8,
     "params_active": 27.8,
     "quantization": "BF16",
     "score_0_100": 44.60873521383075,
     "raw": {
      "metric": "accuracy",
      "value": 0.556869881710646,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.24515588984478145,
     "improvement_points": null,
     "runs": [
      "text:45"
     ],
     "notes": "Open-Jev server, self-hosted"
    },
    {
     "repo": "EldanRing/Winnow-12B",
     "commit": "b2b1421",
     "name": "Winnow-12B",
     "tag": "",
     "model_key": "winnow-12b",
     "maker": "EldanRing",
     "kind": "decision",
     "params_total": 11.91,
     "params_active": 11.91,
     "quantization": "Q8_0",
     "score_0_100": 44.32438580527752,
     "raw": {
      "metric": "accuracy",
      "value": 0.5545950864422202,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1970922224223394,
     "improvement_points": null,
     "runs": [
      "text:128"
     ],
     "notes": "Winnow server (llama.cpp), self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen1",
     "commit": "325378d",
     "name": "KnowLine-9B-Gen1",
     "tag": "",
     "model_key": "knowline-9b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 43.186988171064606,
     "raw": {
      "metric": "accuracy",
      "value": 0.5454959053685169,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.045748292761383084,
     "improvement_points": null,
     "runs": [
      "text:1174"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "apertixai/cerno-9b-v1.2",
     "commit": "f02a632fbe3a4698cd1ff5c0b929ea240e2b8eda",
     "name": "Cerno 9B v1.2",
     "tag": "",
     "model_key": "cerno-9b-v1.2",
     "maker": "Apertix AI",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 43.073248407643305,
     "raw": {
      "metric": "accuracy",
      "value": 0.5445859872611465,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.09795232029117354,
     "improvement_points": null,
     "runs": [
      "text:1201"
     ],
     "notes": "Cerno server (a thin System One server around the card's CernoEngine, transformers: logic's for its English runs, compactor's di_serve.py for its Turkish runs), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen2",
     "commit": "8ec8eb6",
     "name": "KnowLine-9B-Gen2",
     "tag": "",
     "model_key": "knowline-9b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 41.5946314831665,
     "raw": {
      "metric": "accuracy",
      "value": 0.5327570518653321,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.05145641366709974,
     "improvement_points": null,
     "runs": [
      "text:1306"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "newsbang/adjudex-9b-v5",
     "commit": "ffdcb82",
     "name": "Adjudex 9B v5",
     "tag": "",
     "model_key": "adjudex-9b-v5",
     "maker": "newsbang",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 41.42402183803457,
     "raw": {
      "metric": "accuracy",
      "value": 0.5313921747042766,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.09166090025561657,
     "improvement_points": null,
     "runs": [
      "text:1183"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "baselabs/elephant-1",
     "commit": "b3f8a93",
     "name": "Elephant-1",
     "tag": "",
     "model_key": "elephant-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 9.41,
     "params_active": 9.41,
     "quantization": "BF16",
     "score_0_100": 40.45723384895359,
     "raw": {
      "metric": "accuracy",
      "value": 0.5236578707916287,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.11820176478616924,
     "improvement_points": null,
     "runs": [
      "text:373"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-9B",
     "commit": "ab54df5",
     "name": "GLiDE 9B",
     "tag": "",
     "model_key": "glide-9b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 7.94,
     "params_active": 7.94,
     "quantization": "BF16",
     "score_0_100": 40.059144676979066,
     "raw": {
      "metric": "accuracy",
      "value": 0.5204731574158326,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.08527722876180754,
     "improvement_points": null,
     "runs": [
      "text:907"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "Patrick-18/Dexy-Decision-9B-V2",
     "commit": "4f434e2",
     "name": "Dexy-Decision-9B-V2",
     "tag": "",
     "model_key": "dexy-decision-9b-v2",
     "maker": "Patrick-18",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 39.49044585987261,
     "raw": {
      "metric": "accuracy",
      "value": 0.5159235668789809,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.09143110474276758,
     "improvement_points": null,
     "runs": [
      "text:1255"
     ],
     "notes": "Dexy server (the card's Transformers d3_server.py, in its requirements.txt pins: torch 2.14.0, transformers 5.19.0), self-hosted, BF16"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-9B-Instant",
     "commit": "18a17f734966b213b2f00fa9c915e8b9af2a9b20",
     "name": "StartLux-Decision-v1.1-9B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-9b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 39.035486806187436,
     "raw": {
      "metric": "accuracy",
      "value": 0.5122838944494995,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.07621217831446238,
     "improvement_points": null,
     "runs": [
      "text:1290"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.3",
     "commit": "d06fb3f",
     "name": "Perplexity Decider Flash v1.3",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 37.158780709736114,
     "raw": {
      "metric": "accuracy",
      "value": 0.49727024567788897,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1255235993669274,
     "improvement_points": null,
     "runs": [
      "text:1123"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "scar-ai/vera-medium-omni",
     "commit": "1f23a6c",
     "name": "vera-medium-omni",
     "tag": "",
     "model_key": "vera-medium-omni",
     "maker": "scar-ai",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 36.36260236578707,
     "raw": {
      "metric": "accuracy",
      "value": 0.4909008189262966,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03696510464058235,
     "improvement_points": null,
     "runs": [
      "text:1264"
     ],
     "notes": "vera server (the card's own vera-serve from vera-s1 0.1.8, PyTorch engine), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-4B-Gen5.5",
     "commit": "779944d2cc8ffa19cef386aee83c83bcb967ade2",
     "name": "KnowLine 4B Gen5.5",
     "tag": "",
     "model_key": "knowline-4b-gen5.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "FP8 at load",
     "score_0_100": 35.62329390354868,
     "raw": {
      "metric": "accuracy",
      "value": 0.48498635122838946,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.13457633784697456,
     "improvement_points": null,
     "runs": [
      "text:965"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, FP8 at load"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-4B-Instant",
     "commit": "416088f77cfce09633906a3675c7cd2e5dabcf10",
     "name": "StartLux-Decision-v1.1-4B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-4b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 34.82711555959963,
     "raw": {
      "metric": "accuracy",
      "value": 0.4786169244767971,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.07067359984943818,
     "improvement_points": null,
     "runs": [
      "text:1236"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.2",
     "commit": "6d771ad",
     "name": "Perplexity Decider Flash v1.2",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 34.71337579617834,
     "raw": {
      "metric": "accuracy",
      "value": 0.47770700636942676,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.08269965655542914,
     "improvement_points": null,
     "runs": [
      "text:223"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "Valen-Team/Valen-Decision-1.5-9B",
     "commit": "5a51211",
     "name": "Valen-Decision-1.5-9B",
     "tag": "",
     "model_key": "valen-decision-1.5-9b",
     "maker": "Valen-Team",
     "kind": "decision",
     "params_total": 8.4,
     "params_active": 8.4,
     "quantization": "BF16",
     "score_0_100": 33.68971792538671,
     "raw": {
      "metric": "accuracy",
      "value": 0.4695177434030937,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.03768682194780701,
     "improvement_points": null,
     "runs": [
      "text:1275"
     ],
     "notes": "Valen server (the card's valen-vllm plugin, Valen-Team/Valen at 773d3a6, in vllm/vllm-openai v0.33.0), self-hosted, BF16"
    },
    {
     "repo": "victorbennett/Verdin-4B-v1.1",
     "commit": "45830d2",
     "name": "Verdin-4B-v1.1",
     "tag": "",
     "model_key": "verdin-4b-v1.1",
     "maker": "victorbennett",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 33.46223839854412,
     "raw": {
      "metric": "accuracy",
      "value": 0.467697907188353,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.09842375238368434,
     "improvement_points": null,
     "runs": [
      "text:1244"
     ],
     "notes": "Verdin server (the card's own python -m verdin.server, in the package versions its card lists), self-hosted, BF16"
    },
    {
     "repo": "quincyqiang/chinese-decision-pointer-4b",
     "commit": "889f23e",
     "name": "Chinese Decision Pointer 4B",
     "tag": "",
     "model_key": "chinese-decision-pointer-4b",
     "maker": "quincyqiang",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 33.34849863512284,
     "raw": {
      "metric": "accuracy",
      "value": 0.46678798908098273,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.116609689483732,
     "improvement_points": null,
     "runs": [
      "text:1206"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Microsoft-Decision-1",
     "tag": "",
     "model_key": "microsoft-decision-1",
     "maker": "Microsoft",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 33.0641492265696,
     "raw": {
      "metric": "accuracy",
      "value": 0.46451319381255685,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.19786253783505095,
     "improvement_points": null,
     "runs": [
      "text:205"
     ],
     "notes": "OpenRouter System One API"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 32.722929936305725,
     "raw": {
      "metric": "accuracy",
      "value": 0.46178343949044587,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1146"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768"
    },
    {
     "repo": "Cloudflare/clef-flash",
     "commit": "fde727a",
     "name": "Clef-Flash",
     "tag": "",
     "model_key": "clef-flash",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 9.53,
     "params_active": 9.53,
     "quantization": "BF16",
     "score_0_100": 32.154231119199274,
     "raw": {
      "metric": "accuracy",
      "value": 0.4572338489535942,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.15476878980891715,
     "improvement_points": null,
     "runs": [
      "text:139"
     ],
     "notes": "Clef server, self-hosted"
    },
    {
     "repo": "Qwen/Qwen3-14B",
     "commit": "40c0698",
     "name": "Qwen3-14B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3-14b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 32.154231119199274,
     "raw": {
      "metric": "accuracy",
      "value": 0.4572338489535942,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.41002744844910793,
     "improvement_points": null,
     "runs": [
      "text:131"
     ],
     "notes": "vLLM, self-hosted, BF16"
    },
    {
     "repo": "ecloudtech/Erk-14B",
     "commit": "8584b14",
     "name": "Erk-14B",
     "tag": "reasoning off · chat",
     "model_key": "erk-14b",
     "maker": "eCloud Technology",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 31.244313011828933,
     "raw": {
      "metric": "accuracy",
      "value": 0.4499545040946315,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.13455620377261668,
     "improvement_points": null,
     "runs": [
      "text:132"
     ],
     "notes": "vLLM, self-hosted, BF16"
    },
    {
     "repo": "wayfind/metask-jev-4b-policy-mix",
     "commit": "ea20fe8",
     "name": "Metask-Jev-4B",
     "tag": "",
     "model_key": "metask-jev-4b",
     "maker": "Metask Lab",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 25.955414012738853,
     "raw": {
      "metric": "accuracy",
      "value": 0.40764331210191085,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.058126771586986974,
     "improvement_points": null,
     "runs": [
      "text:126"
     ],
     "notes": "Metask-Jev server, self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-2B-Gen1",
     "commit": "53245d470a4ce316e32fd0d18c24e142c697f40b",
     "name": "KnowLine-2B-Gen1",
     "tag": "",
     "model_key": "knowline-2b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 21.007734303912645,
     "raw": {
      "metric": "accuracy",
      "value": 0.3680618744313012,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.08357872663359267,
     "improvement_points": null,
     "runs": [
      "text:1079"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-2B",
     "commit": "00f122c",
     "name": "GLiDE 2B",
     "tag": "",
     "model_key": "glide-2b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 1.88,
     "params_active": 1.88,
     "quantization": "BF16",
     "score_0_100": 19.131028207461327,
     "raw": {
      "metric": "accuracy",
      "value": 0.35304822565969063,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.20675981159556187,
     "improvement_points": null,
     "runs": [
      "text:915"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-2B-Instant",
     "commit": "f1e6cd147dda82c9ffa3e8e725d5c1f175566fb1",
     "name": "StartLux-Decision-v1.1-2B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-2b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 17.823020928116467,
     "raw": {
      "metric": "accuracy",
      "value": 0.34258416742493175,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.06376769947224939,
     "improvement_points": null,
     "runs": [
      "text:977"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "LiquidAI/LFM2.5-2.6B",
     "commit": "7141b56",
     "name": "LFM2.5-2.6B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-2.6b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 2.7,
     "params_active": 2.7,
     "quantization": "BF16",
     "score_0_100": 17.3680618744313,
     "raw": {
      "metric": "accuracy",
      "value": 0.33894449499545043,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1037"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-2.6B@7141b56), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50; output cap 32,768"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning on · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 15.491355777979976,
     "raw": {
      "metric": "accuracy",
      "value": 0.32393084622383983,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1089"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking on; output cap 32,768"
    },
    {
     "repo": "LiquidAI/d1-2B",
     "commit": "10e7671",
     "name": "d1-2B",
     "tag": "",
     "model_key": "liquid-d1-2b",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": 1.92,
     "params_active": 1.92,
     "quantization": "BF16",
     "score_0_100": 11.73794358507734,
     "raw": {
      "metric": "accuracy",
      "value": 0.29390354868061874,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.0579868390338947,
     "improvement_points": null,
     "runs": [
      "text:340"
     ],
     "notes": "d1 server (Liquid's model code in Transformers, the model card's call), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-0.8B",
     "commit": "9eacdee",
     "name": "GLiDE 0.8B",
     "tag": "",
     "model_key": "glide-0.8b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.75,
     "params_active": 0.75,
     "quantization": "BF16",
     "score_0_100": 11.567333939945405,
     "raw": {
      "metric": "accuracy",
      "value": 0.29253867151956325,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.24802735970198903,
     "improvement_points": null,
     "runs": [
      "text:923"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "OmniJev/OneJev1.6-0.8B",
     "commit": "bcd88a382f314b26cb7b1b7f369c1ed3502ee7a0",
     "name": "OneJev1.6-0.8B",
     "tag": "",
     "model_key": "onejev1.6-0.8b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 0.85,
     "params_active": 0.85,
     "quantization": "BF16",
     "score_0_100": 11.396724294813461,
     "raw": {
      "metric": "accuracy",
      "value": 0.2911737943585077,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.09291664285714284,
     "improvement_points": null,
     "runs": [
      "text:1029"
     ],
     "notes": "Flide server (flide 0.1.11, the card's server for OneJev 1.6), self-hosted, BF16"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-0.8B-Instant",
     "commit": "6496ae641cc71b149ce9dad69bd97f2a44e7bf11",
     "name": "StartLux-Decision-v1.1-0.8B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-0.8b-instant",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 10.714285714285712,
     "raw": {
      "metric": "accuracy",
      "value": 0.2857142857142857,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.04396350193467871,
     "improvement_points": null,
     "runs": [
      "text:1015"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning off · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 10.429936305732483,
     "raw": {
      "metric": "accuracy",
      "value": 0.28343949044585987,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1227"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking off; output cap 32,768"
    },
    {
     "repo": "StrandsAgents/strands-decider-2B-hobson-v21",
     "commit": "2b52a62",
     "name": "Strands Decider 2B",
     "tag": "",
     "model_key": "strands-decider-2b",
     "maker": "Strands Agents",
     "kind": "decision",
     "params_total": 2.29,
     "params_active": 2.29,
     "quantization": "BF16",
     "score_0_100": 9.349408553230209,
     "raw": {
      "metric": "accuracy",
      "value": 0.2747952684258417,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.12245345768880803,
     "improvement_points": null,
     "runs": [
      "text:141"
     ],
     "notes": "Strands Decider server, self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-0.8B-Gen1",
     "commit": "6b27831f7b783c03278bd946d4d8f4f7a98a3798",
     "name": "KnowLine-0.8B-Gen1",
     "tag": "",
     "model_key": "knowline-0.8b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 8.894449499545043,
     "raw": {
      "metric": "accuracy",
      "value": 0.27115559599636035,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.08283227420787548,
     "improvement_points": null,
     "runs": [
      "text:1005"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Decider 1",
     "tag": "",
     "model_key": "meragpt-decider-1",
     "maker": "meraGPT",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 8.894449499545043,
     "raw": {
      "metric": "accuracy",
      "value": 0.27115559599636035,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1431224749772521,
     "improvement_points": null,
     "runs": [
      "text:243"
     ],
     "notes": "meraGPT System One API"
    },
    {
     "repo": "Qwen/Qwen3.5-2B",
     "commit": "15852e8",
     "name": "Qwen3.5-2B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-2b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 6.733393994540489,
     "raw": {
      "metric": "accuracy",
      "value": 0.2538671519563239,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1218"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-2B@15852e8), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768"
    },
    {
     "repo": "knowledgator/gliclass-v4-edge",
     "commit": "885af1af8a07edf4ed34df83ca0d995ed8201283",
     "name": "GLiClass v4 Edge",
     "tag": "",
     "model_key": "gliclass-v4-edge",
     "maker": "Knowledgator",
     "kind": "decision",
     "params_total": 0.44,
     "params_active": 0.44,
     "quantization": "BF16",
     "score_0_100": 4.1742493175614195,
     "raw": {
      "metric": "accuracy",
      "value": 0.23339399454049137,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.10511891338772758,
     "improvement_points": null,
     "runs": [
      "text:1046"
     ],
     "notes": "GLiClass server (our servers/gliclass_server.py: the card's gliclass pipeline and its System One mapping), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-0.4B",
     "commit": "6bda7ab",
     "name": "GLiDE 0.4B",
     "tag": "",
     "model_key": "glide-0.4b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.4,
     "params_active": 0.4,
     "quantization": "BF16",
     "score_0_100": 2.8093721565059133,
     "raw": {
      "metric": "accuracy",
      "value": 0.22247497725204732,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.07959284296043859,
     "improvement_points": null,
     "runs": [
      "text:931"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "LiquidAI/LFM2.5-8B-A1B",
     "commit": "dd83f5c",
     "name": "LFM2.5-8B-A1B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-8b-a1b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 8.47,
     "params_active": 1.69,
     "quantization": "BF16",
     "score_0_100": 1.2170154686078245,
     "raw": {
      "metric": "accuracy",
      "value": 0.2097361237488626,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1288"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-8B-A1B@dd83f5c), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.05, seed 0, temperature 0.2, top_k 80; output cap 32,768"
    },
    {
     "repo": "fastino/GLiNER2.5-multi-Decide",
     "commit": "a35a0cd",
     "name": "GLiNER2.5-multi-Decide",
     "tag": "",
     "model_key": "gliner2.5-multi-decide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": 0.25022747952684055,
     "raw": {
      "metric": "accuracy",
      "value": 0.20200181983621474,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.23118755647792288,
     "improvement_points": null,
     "runs": [
      "text:135"
     ],
     "notes": "GLiNER2 server, self-hosted"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya",
     "tag": "",
     "model_key": "laya",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.42,
     "params_active": 0.42,
     "quantization": null,
     "score_0_100": -1.2283894449499553,
     "raw": {
      "metric": "accuracy",
      "value": 0.19017288444040037,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.1321950409463146,
     "improvement_points": null,
     "runs": [
      "text:51"
     ],
     "notes": "Laya server, self-hosted"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya Multilingual",
     "tag": "",
     "model_key": "laya-multilingual",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.32,
     "params_active": 0.32,
     "quantization": null,
     "score_0_100": -2.6501364877161078,
     "raw": {
      "metric": "accuracy",
      "value": 0.17879890809827115,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.2771874886260236,
     "improvement_points": null,
     "runs": [
      "text:46"
     ],
     "notes": "Laya server, self-hosted"
    },
    {
     "repo": "fastino/gliner2.5-multi-v1",
     "commit": "2ca71aa",
     "name": "GLiNER2.5 Multi",
     "tag": "",
     "model_key": "gliner2.5-multi-base",
     "maker": "Fastino",
     "kind": "chat",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": -3.1619654231119196,
     "raw": {
      "metric": "accuracy",
      "value": 0.17470427661510465,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": 0.35619655110491094,
     "improvement_points": null,
     "runs": [
      "text:122"
     ],
     "notes": "GLiNER2 server, self-hosted"
    },
    {
     "repo": "openbmb/MiniCPM5-2B",
     "commit": "42433e9",
     "name": "MiniCPM5-2B",
     "tag": "reasoning off · chat",
     "model_key": "minicpm5-2b",
     "maker": "OpenBMB",
     "kind": "chat",
     "params_total": 2.52,
     "params_active": 2.52,
     "quantization": "BF16",
     "score_0_100": -18.005004549590538,
     "raw": {
      "metric": "accuracy",
      "value": 0.05595996360327571,
      "questions": 2198,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1299"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision openbmb/MiniCPM5-2B@42433e9), one question per request, plain answer line CEVAP: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking off; output cap 32,768"
    }
   ]
  },
  {
   "bench": "turkish-reading",
   "title": "Turkish Reading",
   "url": "https://zekiworks.github.io/analysis/decision-models/",
   "updated": "2026-11-16T20:50:26Z",
   "score": {
    "zero": "chance: 1 of 5 options right on every question (accuracy 20%)",
    "hundred": "every one of the 656 Turkish reading questions (units 1 to 6) right (accuracy 100%)"
   },
   "models": [
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.2",
     "tag": "max · chat",
     "model_key": "claude-fable-5.2",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.14176829268291,
     "raw": {
      "metric": "accuracy",
      "value": 0.9771341463414634,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.08195121951219547,
     "improvement_points": null,
     "runs": [
      "text:980"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Carbon",
     "tag": "high · chat",
     "model_key": "gemini-4-carbon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 96.95121951219512,
     "raw": {
      "metric": "accuracy",
      "value": 0.975609756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.007301829268287369,
     "improvement_points": null,
     "runs": [
      "text:1108"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Hydrogen",
     "tag": "high · chat",
     "model_key": "gemini-4-hydrogen",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 96.5701219512195,
     "raw": {
      "metric": "accuracy",
      "value": 0.9725609756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.023231707317068293,
     "improvement_points": null,
     "runs": [
      "text:1114"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 3.8 Flash",
     "tag": "high · chat",
     "model_key": "gemini-3.8-flash",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 96.3795731707317,
     "raw": {
      "metric": "accuracy",
      "value": 0.9710365853658537,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.013079268292683636,
     "improvement_points": null,
     "runs": [
      "text:140",
      "text:123",
      "text:188"
     ],
     "notes": "Gemini API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Astra",
     "tag": "high · chat",
     "model_key": "gpt-6-astra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 96.3795731707317,
     "raw": {
      "metric": "accuracy",
      "value": 0.9710365853658537,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.009634146341469235,
     "improvement_points": null,
     "runs": [
      "text:277",
      "text:610",
      "text:48",
      "text:191"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Argon",
     "tag": "low · chat",
     "model_key": "gemini-4-argon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 96.1890243902439,
     "raw": {
      "metric": "accuracy",
      "value": 0.9695121951219512,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.010899390243897964,
     "improvement_points": null,
     "runs": [
      "text:557"
     ],
     "notes": "Gemini API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Sol",
     "tag": "high · chat",
     "model_key": "gpt-6.1-sol",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.61737804878048,
     "raw": {
      "metric": "accuracy",
      "value": 0.9649390243902439,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.01798780487805399,
     "improvement_points": null,
     "runs": [
      "text:276",
      "text:124",
      "text:190"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Terra",
     "tag": "high · chat",
     "model_key": "gpt-6.1-terra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.23628048780488,
     "raw": {
      "metric": "accuracy",
      "value": 0.961890243902439,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.013033536585371295,
     "improvement_points": null,
     "runs": [
      "text:275",
      "text:274"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.1",
     "tag": "high · chat",
     "model_key": "claude-fable-5.1",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 94.47408536585367,
     "raw": {
      "metric": "accuracy",
      "value": 0.9557926829268293,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.10850609756097773,
     "improvement_points": null,
     "runs": [
      "text:310"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Opus 5.5",
     "tag": "high · chat",
     "model_key": "claude-opus-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 94.47408536585367,
     "raw": {
      "metric": "accuracy",
      "value": 0.9557926829268293,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.1434908536585367,
     "improvement_points": null,
     "runs": [
      "text:278",
      "text:47",
      "text:192"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Grok 4.7",
     "tag": "high · chat",
     "model_key": "grok-4.7",
     "maker": "xAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 93.71189024390245,
     "raw": {
      "metric": "accuracy",
      "value": 0.9496951219512195,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.105335365853659,
     "improvement_points": null,
     "runs": [
      "text:311"
     ],
     "notes": "xAI Responses API, SuperGrok subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen4-Max",
     "tag": "high · chat",
     "model_key": "qwen4-max",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 93.33079268292683,
     "raw": {
      "metric": "accuracy",
      "value": 0.9466463414634146,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.07010670731707497,
     "improvement_points": null,
     "runs": [
      "text:1336"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint (the maker's), fallbacks off, reasoning effort high, output cap 32,768 tokens, reasoning included; the answer schema"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · 1017",
     "model_key": "deepseek-v4.1-flash-1017",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 92.56859756097559,
     "raw": {
      "metric": "accuracy",
      "value": 0.9405487804878049,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.059530289576940495,
     "improvement_points": null,
     "runs": [
      "text:585",
      "text:558"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; 1017 weights (revision 1bae0e1)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Kimi K3.1 Pro",
     "tag": "low · chat",
     "model_key": "kimi-k3.1-pro",
     "maker": "Moonshot",
     "kind": "chat",
     "params_total": 2800.0,
     "params_active": 104.0,
     "quantization": null,
     "score_0_100": 92.1875,
     "raw": {
      "metric": "accuracy",
      "value": 0.9375,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.043048780487811425,
     "improvement_points": null,
     "runs": [
      "text:320",
      "text:315"
     ],
     "notes": "OpenRouter chat API, Moonshot AI's endpoint (mxfp4), fallbacks off"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · first release",
     "model_key": "deepseek-v4.1-flash-first-release",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": 552.0,
     "params_active": 16.0,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 91.80640243902438,
     "raw": {
      "metric": "accuracy",
      "value": 0.9344512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.06553691607702383,
     "improvement_points": null,
     "runs": [
      "text:121",
      "text:120"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; first-release weights (revision fb2764a), not the 1017 update"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen2",
     "commit": "fd99a9e",
     "name": "KnowLine-31B-Gen2",
     "tag": "",
     "model_key": "knowline-31b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 91.80640243902438,
     "raw": {
      "metric": "accuracy",
      "value": 0.9344512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.06281160798773058,
     "improvement_points": null,
     "runs": [
      "text:1314"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "MachinePulseAI/karpo-decision-31B",
     "commit": "d69cdd5",
     "name": "Karpo Decision 31B",
     "tag": "",
     "model_key": "karpo-decision-31b",
     "maker": "MachinePulse",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 91.80640243902438,
     "raw": {
      "metric": "accuracy",
      "value": 0.9344512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.012282059207248364,
     "improvement_points": null,
     "runs": [
      "text:1156"
     ],
     "notes": "karpo-decision server (the System One server of the card's karpo-decision package, machinepulse-ai/karpo-decision at ac40d39), self-hosted, BF16"
    },
    {
     "repo": "Poindexter-Labs/Copernicus-31B",
     "commit": "be71525",
     "name": "Copernicus-31B",
     "tag": "",
     "model_key": "copernicus-31b",
     "maker": "Poindexter Labs",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 91.42530487804879,
     "raw": {
      "metric": "accuracy",
      "value": 0.9314024390243902,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.06260098268975739,
     "improvement_points": null,
     "runs": [
      "text:1165"
     ],
     "notes": "Copernicus server (the card's copernicus_server.py, Transformers), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen1.5",
     "commit": "8e56aa1b236a6dd1064dcfd22ba1e5e360e4c140",
     "name": "KnowLine-31B-Gen1.5",
     "tag": "",
     "model_key": "knowline-31b-gen1.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 91.04420731707316,
     "raw": {
      "metric": "accuracy",
      "value": 0.9283536585365854,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03369956281247031,
     "improvement_points": null,
     "runs": [
      "text:957"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "turion-ai/turion-jev-31b",
     "commit": "866c0fc6e040cba23fcc72113a80e5c99d5ebab1",
     "name": "Turion Jev 31B",
     "tag": "",
     "model_key": "turion-jev-31b",
     "maker": "Turion.AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 at load",
     "score_0_100": 90.85365853658536,
     "raw": {
      "metric": "accuracy",
      "value": 0.926829268292683,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.09529341615853668,
     "improvement_points": null,
     "runs": [
      "text:1273"
     ],
     "notes": "Turion Jev server (turion_jev.serve on vLLM, the maker's release code), self-hosted, FP8 at load"
    },
    {
     "repo": "decisionmodels/REAX-1.3",
     "commit": "fa4bf7e",
     "name": "REAX 1.3",
     "tag": "",
     "model_key": "reax-1.3",
     "maker": "REAX",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 90.47256097560977,
     "raw": {
      "metric": "accuracy",
      "value": 0.9237804878048781,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.020594747432098777,
     "improvement_points": null,
     "runs": [
      "text:827"
     ],
     "notes": "REAX Engine server (reax13 mode, the maker's release code), self-hosted, FP8"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini Decision 1",
     "tag": "20261111",
     "model_key": "gemini-decision-1",
     "maker": "Google",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 90.09146341463415,
     "raw": {
      "metric": "accuracy",
      "value": 0.9207317073170732,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.027751048761051823,
     "improvement_points": null,
     "runs": [
      "text:1021"
     ],
     "notes": "OpenRouter decisions API, Google AI Studio endpoint"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1-ultra",
     "tag": "",
     "model_key": "liquid-d1-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 90.09146341463415,
     "raw": {
      "metric": "accuracy",
      "value": 0.9207317073170732,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03627395445041045,
     "improvement_points": null,
     "runs": [
      "text:199"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": "logitshq/Logits-Sigma-1-Duo",
     "commit": "a0c1b6e",
     "name": "Logits Sigma-1 Duo",
     "tag": "",
     "model_key": "logits-sigma-1-duo",
     "maker": "Logits",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 89.90091463414635,
     "raw": {
      "metric": "accuracy",
      "value": 0.9192073170731707,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05319305436500552,
     "improvement_points": null,
     "runs": [
      "text:1221"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "sthanika-ai/Sieve-31B-v3",
     "commit": "8fcee858f7f20a50bfa3c8f984c22adb3cbb46a2",
     "name": "Sieve 31B v3",
     "tag": "",
     "model_key": "sieve-31b-v3",
     "maker": "Sthānika AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 89.71036585365854,
     "raw": {
      "metric": "accuracy",
      "value": 0.9176829268292683,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.035043140243901774,
     "improvement_points": null,
     "runs": [
      "text:1241"
     ],
     "notes": "Sieve server (sieve-serve, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.5-397B-A17B",
     "tag": "high · chat",
     "model_key": "qwen-3.5-397b-a17b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 397.0,
     "params_active": 17.0,
     "quantization": null,
     "score_0_100": 88.7576219512195,
     "raw": {
      "metric": "accuracy",
      "value": 0.9100609756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.021676829268298017,
     "improvement_points": null,
     "runs": [
      "text:870"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included"
    },
    {
     "repo": "google/gemma-4-31B-it",
     "commit": "842da37",
     "name": "Gemma 4 31B",
     "tag": "reasoning off · chat",
     "model_key": "gemma-4-31b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 weights and KV cache",
     "score_0_100": 88.5670731707317,
     "raw": {
      "metric": "accuracy",
      "value": 0.9085365853658537,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0807925935006883,
     "improvement_points": null,
     "runs": [
      "text:119",
      "text:189"
     ],
     "notes": "vLLM, self-hosted, FP8"
    },
    {
     "repo": "fastino/GLiDE-27B",
     "commit": "4d32743",
     "name": "GLiDE 27B",
     "tag": "",
     "model_key": "glide-27b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 88.5670731707317,
     "raw": {
      "metric": "accuracy",
      "value": 0.9085365853658537,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.02394488331423235,
     "improvement_points": null,
     "runs": [
      "text:899"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "baselabs/mammoth-1",
     "commit": "61cec44",
     "name": "Mammoth-1",
     "tag": "",
     "model_key": "mammoth-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 87.99542682926827,
     "raw": {
      "metric": "accuracy",
      "value": 0.9039634146341463,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.02013431554877984,
     "improvement_points": null,
     "runs": [
      "text:374"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Muse Ember",
     "tag": "max · chat",
     "model_key": "muse-ember",
     "maker": "Meta",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 87.80487804878048,
     "raw": {
      "metric": "accuracy",
      "value": 0.9024390243902439,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.07487804878048944,
     "improvement_points": null,
     "runs": [
      "text:939"
     ],
     "notes": "OpenRouter chat API, Meta's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning max"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2-ultra",
     "tag": "",
     "model_key": "liquid-d1.2-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 87.23323170731706,
     "raw": {
      "metric": "accuracy",
      "value": 0.8978658536585366,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05639812057083431,
     "improvement_points": null,
     "runs": [
      "text:1347"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": "vllm-sr/d3-ultra",
     "commit": "52ba9e7",
     "name": "d3-ultra",
     "tag": "",
     "model_key": "vllm-sr-d3-ultra",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 87.23323170731706,
     "raw": {
      "metric": "accuracy",
      "value": 0.8978658536585366,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03791494886497946,
     "improvement_points": null,
     "runs": [
      "text:269"
     ],
     "notes": "d3 package server (transformers, the release code), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE",
     "commit": "a7e4589",
     "name": "GLiDE",
     "tag": "v1.2",
     "model_key": "fastino-glide-v1.2",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 87.04268292682926,
     "raw": {
      "metric": "accuracy",
      "value": 0.8963414634146342,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03469532691718118,
     "improvement_points": null,
     "runs": [
      "text:248"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "nace-ai/drex-max-v1",
     "commit": "cac65e1",
     "name": "Drex Max",
     "tag": "",
     "model_key": "drex-max-v1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 87.04268292682926,
     "raw": {
      "metric": "accuracy",
      "value": 0.8963414634146342,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.034324895352650375,
     "improvement_points": null,
     "runs": [
      "text:322"
     ],
     "notes": "Drex Max server (Nace's loader, drex_max), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3-Flash",
     "tag": "reasoning on · chat",
     "model_key": "glm-5.3-flash",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 320.0,
     "params_active": 18.0,
     "quantization": "EXL3 4-bit experts, FP8 dense layers and KV cache",
     "score_0_100": 86.66158536585367,
     "raw": {
      "metric": "accuracy",
      "value": 0.8932926829268293,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.054695121951221254,
     "improvement_points": null,
     "runs": [
      "text:160",
      "text:159"
     ],
     "notes": "TensorFold, self-hosted, EXL3 4-bit experts"
    },
    {
     "repo": "Qwen/Qwen3.8-27B",
     "commit": "1d4bf0f",
     "name": "Qwen3.8-27B",
     "tag": "low · chat",
     "model_key": "qwen3.8-27b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 27.78,
     "params_active": 27.78,
     "quantization": "BF16",
     "score_0_100": 86.66158536585367,
     "raw": {
      "metric": "accuracy",
      "value": 0.8932926829268293,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.10664740078379775,
     "improvement_points": null,
     "runs": [
      "text:318",
      "text:50",
      "text:53",
      "text:83"
     ],
     "notes": "vLLM, self-hosted"
    },
    {
     "repo": "vllm-sr/d3",
     "commit": "20f37b9",
     "name": "d3",
     "tag": "",
     "model_key": "vllm-sr-d3",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 86.47103658536584,
     "raw": {
      "metric": "accuracy",
      "value": 0.8917682926829268,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03512189413933619,
     "improvement_points": null,
     "runs": [
      "text:259"
     ],
     "notes": "vLLM System One endpoint (nightly), self-hosted, BF16"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.2",
     "commit": "225c913",
     "name": "Perplexity Decider 27B v1.2",
     "tag": "DGX Spark",
     "model_key": "perplexity-decider-27b-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 86.47103658536584,
     "raw": {
      "metric": "accuracy",
      "value": 0.8917682926829268,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03089436055115348,
     "improvement_points": null,
     "runs": [
      "text:217",
      "text:633"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.3",
     "commit": "9c2ebb2",
     "name": "Perplexity Decider 27B v1.3",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 86.28048780487804,
     "raw": {
      "metric": "accuracy",
      "value": 0.8902439024390244,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.023541443402550113,
     "improvement_points": null,
     "runs": [
      "text:1134"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 86.28048780487804,
     "raw": {
      "metric": "accuracy",
      "value": 0.8902439024390244,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.030055772126997168,
     "improvement_points": null,
     "runs": [
      "text:145"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.2",
     "tag": "API · with web",
     "model_key": "perplexity-decider-27b-v1.2-with-web",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 86.08993902439023,
     "raw": {
      "metric": "accuracy",
      "value": 0.8887195121951219,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.027289047563405975,
     "improvement_points": null,
     "runs": [
      "text:640"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex Max v1.1",
     "tag": "",
     "model_key": "drex-max-1.1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 85.70884146341461,
     "raw": {
      "metric": "accuracy",
      "value": 0.885670731707317,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.033872135270565616,
     "improvement_points": null,
     "runs": [
      "text:326"
     ],
     "notes": "Nace.AI System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLiDE",
     "tag": "API",
     "model_key": "fastino-glide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 84.75609756097559,
     "raw": {
      "metric": "accuracy",
      "value": 0.8780487804878049,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05800380060037645,
     "improvement_points": null,
     "runs": [
      "text:129"
     ],
     "notes": "Fastino System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3",
     "tag": "low · chat",
     "model_key": "glm-5.3",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 753.33,
     "params_active": 753.33,
     "quantization": null,
     "score_0_100": 84.5655487804878,
     "raw": {
      "metric": "accuracy",
      "value": 0.8765243902439024,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.08608231707317021,
     "improvement_points": null,
     "runs": [
      "text:314"
     ],
     "notes": "OpenRouter chat API, Fireworks endpoint, fallbacks off"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap Preview",
     "tag": "",
     "model_key": "subq-snap-preview",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 84.5655487804878,
     "raw": {
      "metric": "accuracy",
      "value": 0.8765243902439024,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.04234108156183871,
     "improvement_points": null,
     "runs": [
      "text:177"
     ],
     "notes": "SubQ Decisions API (public preview)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "low · chat",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 84.1844512195122,
     "raw": {
      "metric": "accuracy",
      "value": 0.8734756097560976,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.08989329268293247,
     "improvement_points": null,
     "runs": [
      "text:133"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Inkling-Small",
     "tag": "high · chat · Jul 2026",
     "model_key": "inkling-small",
     "maker": "Thinking Machines",
     "kind": "chat",
     "params_total": 265.96,
     "params_active": 12.0,
     "quantization": null,
     "score_0_100": 83.99390243902438,
     "raw": {
      "metric": "accuracy",
      "value": 0.8719512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03228658536585251,
     "improvement_points": null,
     "runs": [
      "text:866"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included; no response format (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.8 Flash",
     "tag": "no reasoning · chat",
     "model_key": "qwen3.8-flash",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 83.99390243902438,
     "raw": {
      "metric": "accuracy",
      "value": 0.8719512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05728658536585157,
     "improvement_points": null,
     "runs": [
      "text:985"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint, fallbacks off, reasoning off"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap 1.1",
     "tag": "",
     "model_key": "subq-snap-1.1",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 83.80335365853658,
     "raw": {
      "metric": "accuracy",
      "value": 0.8704268292682927,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03613463557529743,
     "improvement_points": null,
     "runs": [
      "text:346"
     ],
     "notes": "SubQ Decisions API (public preview)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Haiku 5.5",
     "tag": "max · chat",
     "model_key": "claude-haiku-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 83.61280487804879,
     "raw": {
      "metric": "accuracy",
      "value": 0.8689024390243902,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.189070512820513,
     "improvement_points": null,
     "runs": [
      "text:948"
     ],
     "notes": "OpenRouter chat API, Anthropic's own endpoint, fallbacks off, output cap 128,000 tokens, reasoning max"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiMo-V2.6-Flash",
     "tag": "high · chat",
     "model_key": "mimo-v2.6-flash",
     "maker": "Xiaomi",
     "kind": "chat",
     "params_total": 309.0,
     "params_active": 15.0,
     "quantization": null,
     "score_0_100": 83.23170731707316,
     "raw": {
      "metric": "accuracy",
      "value": 0.8658536585365854,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.037312977099238455,
     "improvement_points": null,
     "runs": [
      "text:877"
     ],
     "notes": "OpenRouter chat API, Xiaomi's own endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included"
    },
    {
     "repo": "PelaAI/KnowLine-12B-Gen1",
     "commit": "f26fbc2",
     "name": "KnowLine-12B-Gen1",
     "tag": "",
     "model_key": "knowline-12b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 82.46951219512194,
     "raw": {
      "metric": "accuracy",
      "value": 0.8597560975609756,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.11431740693505503,
     "improvement_points": null,
     "runs": [
      "text:1309"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2",
     "tag": "",
     "model_key": "liquid-d1.2",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.46951219512194,
     "raw": {
      "metric": "accuracy",
      "value": 0.8597560975609756,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.06017090221918124,
     "improvement_points": null,
     "runs": [
      "text:1346"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 82.46951219512194,
     "raw": {
      "metric": "accuracy",
      "value": 0.8597560975609756,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.04954557947264652,
     "improvement_points": null,
     "runs": [
      "text:44"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "Decisions API",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 81.89786585365854,
     "raw": {
      "metric": "accuracy",
      "value": 0.8551829268292683,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0376641221374054,
     "improvement_points": null,
     "runs": [
      "text:143"
     ],
     "notes": "OpenAI Decisions API (public beta)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Sonnet 5.5",
     "tag": "high · chat",
     "model_key": "claude-sonnet-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 81.51676829268291,
     "raw": {
      "metric": "accuracy",
      "value": 0.8521341463414634,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.13841463414634156,
     "improvement_points": null,
     "runs": [
      "text:55",
      "text:193",
      "text:54"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "celeris-1-decision",
     "tag": "",
     "model_key": "celeris-1-decision",
     "maker": "Celeris",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 81.51676829268291,
     "raw": {
      "metric": "accuracy",
      "value": 0.8521341463414634,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.06225609756097596,
     "improvement_points": null,
     "runs": [
      "text:150"
     ],
     "notes": "Celeris System One API"
    },
    {
     "repo": "OmniJev/OneJev-27B",
     "commit": "a5e5f55",
     "name": "OneJev-27B",
     "tag": "",
     "model_key": "onejev-27b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 81.51676829268291,
     "raw": {
      "metric": "accuracy",
      "value": 0.8521341463414634,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.09873919054878043,
     "improvement_points": null,
     "runs": [
      "text:264"
     ],
     "notes": "OneJev server (qev, OmniJev's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.14",
     "tag": "",
     "model_key": "jev-1.14.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 81.32621951219512,
     "raw": {
      "metric": "accuracy",
      "value": 0.850609756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.021783536585365134,
     "improvement_points": null,
     "runs": [
      "text:194"
     ],
     "notes": "TypeSafe System One API"
    },
    {
     "repo": "Cloudflare/clef",
     "commit": "2f3de3d",
     "name": "Clef",
     "tag": "",
     "model_key": "clef",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 27.48,
     "params_active": 27.48,
     "quantization": "BF16",
     "score_0_100": 81.32621951219512,
     "raw": {
      "metric": "accuracy",
      "value": 0.850609756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0400698170731706,
     "improvement_points": null,
     "runs": [
      "text:125"
     ],
     "notes": "Clef server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.1",
     "tag": "",
     "model_key": "liquid-d1.1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 80.7545731707317,
     "raw": {
      "metric": "accuracy",
      "value": 0.8460365853658537,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0417706768645985,
     "improvement_points": null,
     "runs": [
      "text:254"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Mercury Decide",
     "tag": "",
     "model_key": "inception-mercury-decide",
     "maker": "Inception",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 80.5640243902439,
     "raw": {
      "metric": "accuracy",
      "value": 0.8445121951219512,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.07464567170004002,
     "improvement_points": null,
     "runs": [
      "text:204"
     ],
     "notes": "OpenRouter System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.13",
     "tag": "",
     "model_key": "jev-1.13.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 79.61128048780488,
     "raw": {
      "metric": "accuracy",
      "value": 0.836890243902439,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03803353658536662,
     "improvement_points": null,
     "runs": [
      "text:49"
     ],
     "notes": "TypeSafe System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiniMax-M3.1-Flash",
     "tag": "high · chat",
     "model_key": "minimax-m3.1-flash",
     "maker": "MiniMax",
     "kind": "chat",
     "params_total": 428.0,
     "params_active": 23.0,
     "quantization": null,
     "score_0_100": 78.65853658536584,
     "raw": {
      "metric": "accuracy",
      "value": 0.8292682926829268,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0947713414634172,
     "improvement_points": null,
     "runs": [
      "text:862"
     ],
     "notes": "OpenRouter chat API, MiniMax's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning included; JSON mode without the answer schema (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": "tattrongvu/Jeffan-9B-1211",
     "commit": "862d1a7",
     "name": "Jeffan-9B-1211",
     "tag": "",
     "model_key": "jeffan-9b-1211",
     "maker": "tattrongvu",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 78.65853658536584,
     "raw": {
      "metric": "accuracy",
      "value": 0.8292682926829268,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.11381959456970998,
     "improvement_points": null,
     "runs": [
      "text:1192"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "ZefanCai/Open-Jev-27B-v1.1",
     "commit": "28cf730",
     "name": "Open-Jev-27B-v1.1",
     "tag": "",
     "model_key": "open-jev-27b-v1.1",
     "maker": "Zefan Cai",
     "kind": "decision",
     "params_total": 27.8,
     "params_active": 27.8,
     "quantization": "BF16",
     "score_0_100": 78.65853658536584,
     "raw": {
      "metric": "accuracy",
      "value": 0.8292682926829268,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.4356508427406811,
     "improvement_points": null,
     "runs": [
      "text:45"
     ],
     "notes": "Open-Jev server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1",
     "tag": "",
     "model_key": "liquid-d1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 78.46798780487804,
     "raw": {
      "metric": "accuracy",
      "value": 0.8277439024390244,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.028235627971199877,
     "improvement_points": null,
     "runs": [
      "text:52"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Cygnet",
     "tag": "",
     "model_key": "cygnet",
     "maker": "Blockbrain Labs",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 78.46798780487804,
     "raw": {
      "metric": "accuracy",
      "value": 0.8277439024390244,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.04132135754946696,
     "improvement_points": null,
     "runs": [
      "text:127"
     ],
     "notes": "Cygnet decision server on vLLM, self-hosted"
    },
    {
     "repo": "EldanRing/Winnow-12B",
     "commit": "b2b1421",
     "name": "Winnow-12B",
     "tag": "",
     "model_key": "winnow-12b",
     "maker": "EldanRing",
     "kind": "decision",
     "params_total": 11.91,
     "params_active": 11.91,
     "quantization": "Q8_0",
     "score_0_100": 77.51524390243902,
     "raw": {
      "metric": "accuracy",
      "value": 0.8201219512195121,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.09320697631220495,
     "improvement_points": null,
     "runs": [
      "text:128"
     ],
     "notes": "Winnow server (llama.cpp), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex 2.0",
     "tag": "",
     "model_key": "drex-2.0",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 35.11,
     "params_active": 3.0,
     "quantization": null,
     "score_0_100": 76.7530487804878,
     "raw": {
      "metric": "accuracy",
      "value": 0.8140243902439024,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.10062500000000009,
     "improvement_points": null,
     "runs": [
      "text:334"
     ],
     "notes": "Nace.AI System One API"
    },
    {
     "repo": "newsbang/adjudex-9b-v5",
     "commit": "ffdcb82",
     "name": "Adjudex 9B v5",
     "tag": "",
     "model_key": "adjudex-9b-v5",
     "maker": "newsbang",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 75.99085365853658,
     "raw": {
      "metric": "accuracy",
      "value": 0.8079268292682927,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.025417447322248297,
     "improvement_points": null,
     "runs": [
      "text:1183"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "baselabs/elephant-1",
     "commit": "b3f8a93",
     "name": "Elephant-1",
     "tag": "",
     "model_key": "elephant-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 9.41,
     "params_active": 9.41,
     "quantization": "BF16",
     "score_0_100": 75.60975609756099,
     "raw": {
      "metric": "accuracy",
      "value": 0.8048780487804879,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.04277260823170779,
     "improvement_points": null,
     "runs": [
      "text:373"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-9B",
     "commit": "ab54df5",
     "name": "GLiDE 9B",
     "tag": "",
     "model_key": "glide-9b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 7.94,
     "params_active": 7.94,
     "quantization": "BF16",
     "score_0_100": 75.22865853658536,
     "raw": {
      "metric": "accuracy",
      "value": 0.801829268292683,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03741752052743735,
     "improvement_points": null,
     "runs": [
      "text:907"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen1",
     "commit": "325378d",
     "name": "KnowLine-9B-Gen1",
     "tag": "",
     "model_key": "knowline-9b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 74.84756097560977,
     "raw": {
      "metric": "accuracy",
      "value": 0.7987804878048781,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.017336468117365566,
     "improvement_points": null,
     "runs": [
      "text:1174"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "apertixai/cerno-9b-v1.2",
     "commit": "f02a632fbe3a4698cd1ff5c0b929ea240e2b8eda",
     "name": "Cerno 9B v1.2",
     "tag": "",
     "model_key": "cerno-9b-v1.2",
     "maker": "Apertix AI",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 74.65701219512194,
     "raw": {
      "metric": "accuracy",
      "value": 0.7972560975609756,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03021341463414638,
     "improvement_points": null,
     "runs": [
      "text:1201"
     ],
     "notes": "Cerno server (a thin System One server around the card's CernoEngine, transformers: logic's for its English runs, compactor's di_serve.py for its Turkish runs), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen2",
     "commit": "8ec8eb6",
     "name": "KnowLine-9B-Gen2",
     "tag": "",
     "model_key": "knowline-9b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 74.46646341463415,
     "raw": {
      "metric": "accuracy",
      "value": 0.7957317073170732,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.08237542872976275,
     "improvement_points": null,
     "runs": [
      "text:1306"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "Patrick-18/Dexy-Decision-9B-V2",
     "commit": "4f434e2",
     "name": "Dexy-Decision-9B-V2",
     "tag": "",
     "model_key": "dexy-decision-9b-v2",
     "maker": "Patrick-18",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 74.46646341463415,
     "raw": {
      "metric": "accuracy",
      "value": 0.7957317073170732,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.032350203055034936,
     "improvement_points": null,
     "runs": [
      "text:1255"
     ],
     "notes": "Dexy server (the card's Transformers d3_server.py, in its requirements.txt pins: torch 2.14.0, transformers 5.19.0), self-hosted, BF16"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning on · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 73.89481707317071,
     "raw": {
      "metric": "accuracy",
      "value": 0.7911585365853658,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1149"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking on; output cap 32,768"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-9B-Instant",
     "commit": "18a17f734966b213b2f00fa9c915e8b9af2a9b20",
     "name": "StartLux-Decision-v1.1-9B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-9b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 73.51371951219512,
     "raw": {
      "metric": "accuracy",
      "value": 0.788109756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03587679412996808,
     "improvement_points": null,
     "runs": [
      "text:1290"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "google/gemma-4-E4B-it",
     "commit": "4ecc2ba",
     "name": "Gemma 4 E4B",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-e4b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 8.0,
     "params_active": 8.0,
     "quantization": "BF16",
     "score_0_100": 71.98932926829268,
     "raw": {
      "metric": "accuracy",
      "value": 0.7759146341463414,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1040"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E4B-it@4ecc2ba), one question per request, plain answer line CEVAP: X (no answer schema); sampling temperature 0.0; output cap 8,192"
    },
    {
     "repo": "scar-ai/vera-medium-omni",
     "commit": "1f23a6c",
     "name": "vera-medium-omni",
     "tag": "",
     "model_key": "vera-medium-omni",
     "maker": "scar-ai",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 69.32164634146339,
     "raw": {
      "metric": "accuracy",
      "value": 0.7545731707317073,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.09082240853658538,
     "improvement_points": null,
     "runs": [
      "text:1264"
     ],
     "notes": "vera server (the card's own vera-serve from vera-s1 0.1.8, PyTorch engine), self-hosted, BF16"
    },
    {
     "repo": "Valen-Team/Valen-Decision-1.5-9B",
     "commit": "5a51211",
     "name": "Valen-Decision-1.5-9B",
     "tag": "",
     "model_key": "valen-decision-1.5-9b",
     "maker": "Valen-Team",
     "kind": "decision",
     "params_total": 8.4,
     "params_active": 8.4,
     "quantization": "BF16",
     "score_0_100": 68.75,
     "raw": {
      "metric": "accuracy",
      "value": 0.75,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.09656381527866759,
     "improvement_points": null,
     "runs": [
      "text:1275"
     ],
     "notes": "Valen server (the card's valen-vllm plugin, Valen-Team/Valen at 773d3a6, in vllm/vllm-openai v0.33.0), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-4B-Gen5.5",
     "commit": "779944d2cc8ffa19cef386aee83c83bcb967ade2",
     "name": "KnowLine 4B Gen5.5",
     "tag": "",
     "model_key": "knowline-4b-gen5.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "FP8 at load",
     "score_0_100": 68.5594512195122,
     "raw": {
      "metric": "accuracy",
      "value": 0.7484756097560976,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.06189619612555461,
     "improvement_points": null,
     "runs": [
      "text:965"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, FP8 at load"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.3",
     "commit": "d06fb3f",
     "name": "Perplexity Decider Flash v1.3",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 67.98780487804879,
     "raw": {
      "metric": "accuracy",
      "value": 0.7439024390243902,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.062095331149012355,
     "improvement_points": null,
     "runs": [
      "text:1123"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-4B-Instant",
     "commit": "416088f77cfce09633906a3675c7cd2e5dabcf10",
     "name": "StartLux-Decision-v1.1-4B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-4b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 67.60670731707316,
     "raw": {
      "metric": "accuracy",
      "value": 0.7408536585365854,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.025668962084869594,
     "improvement_points": null,
     "runs": [
      "text:1236"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Microsoft-Decision-1",
     "tag": "",
     "model_key": "microsoft-decision-1",
     "maker": "Microsoft",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 66.84451219512194,
     "raw": {
      "metric": "accuracy",
      "value": 0.7347560975609756,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.08874299271483316,
     "improvement_points": null,
     "runs": [
      "text:205"
     ],
     "notes": "OpenRouter System One API"
    },
    {
     "repo": "Cloudflare/clef-flash",
     "commit": "fde727a",
     "name": "Clef-Flash",
     "tag": "",
     "model_key": "clef-flash",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 9.53,
     "params_active": 9.53,
     "quantization": "BF16",
     "score_0_100": 66.27286585365854,
     "raw": {
      "metric": "accuracy",
      "value": 0.7301829268292683,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05586707317073188,
     "improvement_points": null,
     "runs": [
      "text:139"
     ],
     "notes": "Clef server, self-hosted"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.2",
     "commit": "6d771ad",
     "name": "Perplexity Decider Flash v1.2",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 65.89176829268291,
     "raw": {
      "metric": "accuracy",
      "value": 0.7271341463414634,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0373797512107109,
     "improvement_points": null,
     "runs": [
      "text:223"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "ecloudtech/Erk-14B",
     "commit": "8584b14",
     "name": "Erk-14B",
     "tag": "reasoning off · chat",
     "model_key": "erk-14b",
     "maker": "eCloud Technology",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 65.3201219512195,
     "raw": {
      "metric": "accuracy",
      "value": 0.7225609756097561,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0671924244386414,
     "improvement_points": null,
     "runs": [
      "text:132"
     ],
     "notes": "vLLM, self-hosted, BF16"
    },
    {
     "repo": "Qwen/Qwen3-14B",
     "commit": "40c0698",
     "name": "Qwen3-14B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3-14b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 65.1295731707317,
     "raw": {
      "metric": "accuracy",
      "value": 0.7210365853658537,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.23195079176313713,
     "improvement_points": null,
     "runs": [
      "text:131"
     ],
     "notes": "vLLM, self-hosted, BF16"
    },
    {
     "repo": "victorbennett/Verdin-4B-v1.1",
     "commit": "45830d2",
     "name": "Verdin-4B-v1.1",
     "tag": "",
     "model_key": "verdin-4b-v1.1",
     "maker": "victorbennett",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 63.98628048780488,
     "raw": {
      "metric": "accuracy",
      "value": 0.711890243902439,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.038603305707376225,
     "improvement_points": null,
     "runs": [
      "text:1244"
     ],
     "notes": "Verdin server (the card's own python -m verdin.server, in the package versions its card lists), self-hosted, BF16"
    },
    {
     "repo": "quincyqiang/chinese-decision-pointer-4b",
     "commit": "889f23e",
     "name": "Chinese Decision Pointer 4B",
     "tag": "",
     "model_key": "chinese-decision-pointer-4b",
     "maker": "quincyqiang",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 63.605182926829265,
     "raw": {
      "metric": "accuracy",
      "value": 0.7088414634146342,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0327290906905249,
     "improvement_points": null,
     "runs": [
      "text:1206"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 60.55640243902438,
     "raw": {
      "metric": "accuracy",
      "value": 0.6844512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1146"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768"
    },
    {
     "repo": "wayfind/metask-jev-4b-policy-mix",
     "commit": "ea20fe8",
     "name": "Metask-Jev-4B",
     "tag": "",
     "model_key": "metask-jev-4b",
     "maker": "Metask Lab",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 53.506097560975604,
     "raw": {
      "metric": "accuracy",
      "value": 0.6280487804878049,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05316175708956106,
     "improvement_points": null,
     "runs": [
      "text:126"
     ],
     "notes": "Metask-Jev server, self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-2B-Gen1",
     "commit": "53245d470a4ce316e32fd0d18c24e142c697f40b",
     "name": "KnowLine-2B-Gen1",
     "tag": "",
     "model_key": "knowline-2b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 42.83536585365854,
     "raw": {
      "metric": "accuracy",
      "value": 0.5426829268292683,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.019317639843370166,
     "improvement_points": null,
     "runs": [
      "text:1079"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-2B-Instant",
     "commit": "f1e6cd147dda82c9ffa3e8e725d5c1f175566fb1",
     "name": "StartLux-Decision-v1.1-2B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-2b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 41.12042682926828,
     "raw": {
      "metric": "accuracy",
      "value": 0.5289634146341463,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.03293383635798188,
     "improvement_points": null,
     "runs": [
      "text:977"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-2B",
     "commit": "00f122c",
     "name": "GLiDE 2B",
     "tag": "",
     "model_key": "glide-2b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 1.88,
     "params_active": 1.88,
     "quantization": "BF16",
     "score_0_100": 39.02439024390244,
     "raw": {
      "metric": "accuracy",
      "value": 0.5121951219512195,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.16496506364752234,
     "improvement_points": null,
     "runs": [
      "text:915"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "LiquidAI/LFM2.5-2.6B",
     "commit": "7141b56",
     "name": "LFM2.5-2.6B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-2.6b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 2.7,
     "params_active": 2.7,
     "quantization": "BF16",
     "score_0_100": 31.78353658536585,
     "raw": {
      "metric": "accuracy",
      "value": 0.45426829268292684,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1037"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-2.6B@7141b56), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50; output cap 32,768"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning on · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 30.25914634146341,
     "raw": {
      "metric": "accuracy",
      "value": 0.4420731707317073,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1089"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking on; output cap 32,768"
    },
    {
     "repo": "fastino/GLiDE-0.8B",
     "commit": "9eacdee",
     "name": "GLiDE 0.8B",
     "tag": "",
     "model_key": "glide-0.8b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.75,
     "params_active": 0.75,
     "quantization": "BF16",
     "score_0_100": 27.97256097560975,
     "raw": {
      "metric": "accuracy",
      "value": 0.42378048780487804,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.2182240433769342,
     "improvement_points": null,
     "runs": [
      "text:923"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "LiquidAI/d1-2B",
     "commit": "10e7671",
     "name": "d1-2B",
     "tag": "",
     "model_key": "liquid-d1-2b",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": 1.92,
     "params_active": 1.92,
     "quantization": "BF16",
     "score_0_100": 26.067073170731703,
     "raw": {
      "metric": "accuracy",
      "value": 0.40853658536585363,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.04391152393393037,
     "improvement_points": null,
     "runs": [
      "text:340"
     ],
     "notes": "d1 server (Liquid's model code in Transformers, the model card's call), self-hosted, BF16"
    },
    {
     "repo": "OmniJev/OneJev1.6-0.8B",
     "commit": "bcd88a382f314b26cb7b1b7f369c1ed3502ee7a0",
     "name": "OneJev1.6-0.8B",
     "tag": "",
     "model_key": "onejev1.6-0.8b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 0.85,
     "params_active": 0.85,
     "quantization": "BF16",
     "score_0_100": 24.54268292682927,
     "raw": {
      "metric": "accuracy",
      "value": 0.39634146341463417,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.0734142225609756,
     "improvement_points": null,
     "runs": [
      "text:1029"
     ],
     "notes": "Flide server (flide 0.1.11, the card's server for OneJev 1.6), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-0.8B-Gen1",
     "commit": "6b27831f7b783c03278bd946d4d8f4f7a98a3798",
     "name": "KnowLine-0.8B-Gen1",
     "tag": "",
     "model_key": "knowline-0.8b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 22.827743902439025,
     "raw": {
      "metric": "accuracy",
      "value": 0.3826219512195122,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.05530760715463055,
     "improvement_points": null,
     "runs": [
      "text:1005"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning off · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 21.49390243902439,
     "raw": {
      "metric": "accuracy",
      "value": 0.3719512195121951,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1227"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking off; output cap 32,768"
    },
    {
     "repo": "StrandsAgents/strands-decider-2B-hobson-v21",
     "commit": "2b52a62",
     "name": "Strands Decider 2B",
     "tag": "",
     "model_key": "strands-decider-2b",
     "maker": "Strands Agents",
     "kind": "decision",
     "params_total": 2.29,
     "params_active": 2.29,
     "quantization": "BF16",
     "score_0_100": 21.112804878048777,
     "raw": {
      "metric": "accuracy",
      "value": 0.36890243902439024,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.10225609756097565,
     "improvement_points": null,
     "runs": [
      "text:141"
     ],
     "notes": "Strands Decider server, self-hosted"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-0.8B-Instant",
     "commit": "6496ae641cc71b149ce9dad69bd97f2a44e7bf11",
     "name": "StartLux-Decision-v1.1-0.8B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-0.8b-instant",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 18.826219512195117,
     "raw": {
      "metric": "accuracy",
      "value": 0.35060975609756095,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.050175267023180044,
     "improvement_points": null,
     "runs": [
      "text:1015"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Decider 1",
     "tag": "",
     "model_key": "meragpt-decider-1",
     "maker": "meraGPT",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 18.254573170731703,
     "raw": {
      "metric": "accuracy",
      "value": 0.34603658536585363,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.11790442073170729,
     "improvement_points": null,
     "runs": [
      "text:243"
     ],
     "notes": "meraGPT System One API"
    },
    {
     "repo": "Qwen/Qwen3.5-2B",
     "commit": "15852e8",
     "name": "Qwen3.5-2B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-2b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 17.8734756097561,
     "raw": {
      "metric": "accuracy",
      "value": 0.3429878048780488,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1218"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-2B@15852e8), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768"
    },
    {
     "repo": "knowledgator/gliclass-v4-edge",
     "commit": "885af1af8a07edf4ed34df83ca0d995ed8201283",
     "name": "GLiClass v4 Edge",
     "tag": "",
     "model_key": "gliclass-v4-edge",
     "maker": "Knowledgator",
     "kind": "decision",
     "params_total": 0.44,
     "params_active": 0.44,
     "quantization": "BF16",
     "score_0_100": 7.393292682926829,
     "raw": {
      "metric": "accuracy",
      "value": 0.25914634146341464,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.11856716221020526,
     "improvement_points": null,
     "runs": [
      "text:1046"
     ],
     "notes": "GLiClass server (our servers/gliclass_server.py: the card's gliclass pipeline and its System One mapping), self-hosted, BF16"
    },
    {
     "repo": "LiquidAI/LFM2.5-8B-A1B",
     "commit": "dd83f5c",
     "name": "LFM2.5-8B-A1B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-8b-a1b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 8.47,
     "params_active": 1.69,
     "quantization": "BF16",
     "score_0_100": 7.012195121951219,
     "raw": {
      "metric": "accuracy",
      "value": 0.25609756097560976,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1288"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-8B-A1B@dd83f5c), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.05, seed 0, temperature 0.2, top_k 80; output cap 32,768"
    },
    {
     "repo": "fastino/GLiDE-0.4B",
     "commit": "6bda7ab",
     "name": "GLiDE 0.4B",
     "tag": "",
     "model_key": "glide-0.4b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.4,
     "params_active": 0.4,
     "quantization": "BF16",
     "score_0_100": 5.678353658536583,
     "raw": {
      "metric": "accuracy",
      "value": 0.24542682926829268,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.07235010849529888,
     "improvement_points": null,
     "runs": [
      "text:931"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "fastino/GLiNER2.5-multi-Decide",
     "commit": "a35a0cd",
     "name": "GLiNER2.5-multi-Decide",
     "tag": "",
     "model_key": "gliner2.5-multi-decide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": 1.2957317073170709,
     "raw": {
      "metric": "accuracy",
      "value": 0.21036585365853658,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.2412625576664762,
     "improvement_points": null,
     "runs": [
      "text:135"
     ],
     "notes": "GLiNER2 server, self-hosted"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya",
     "tag": "",
     "model_key": "laya",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.42,
     "params_active": 0.42,
     "quantization": null,
     "score_0_100": 1.1051829268292659,
     "raw": {
      "metric": "accuracy",
      "value": 0.20884146341463414,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.09224176829268296,
     "improvement_points": null,
     "runs": [
      "text:51"
     ],
     "notes": "Laya server, self-hosted"
    },
    {
     "repo": "fastino/gliner2.5-multi-v1",
     "commit": "2ca71aa",
     "name": "GLiNER2.5 Multi",
     "tag": "",
     "model_key": "gliner2.5-multi-base",
     "maker": "Fastino",
     "kind": "chat",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": -3.4679878048780517,
     "raw": {
      "metric": "accuracy",
      "value": 0.1722560975609756,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.3512977628582498,
     "improvement_points": null,
     "runs": [
      "text:122"
     ],
     "notes": "GLiNER2 server, self-hosted"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya Multilingual",
     "tag": "",
     "model_key": "laya-multilingual",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.32,
     "params_active": 0.32,
     "quantization": null,
     "score_0_100": -3.6585365853658534,
     "raw": {
      "metric": "accuracy",
      "value": 0.17073170731707318,
      "questions": 656,
      "options": 5
     },
     "calibration_error": 0.33111707317073164,
     "improvement_points": null,
     "runs": [
      "text:46"
     ],
     "notes": "Laya server, self-hosted"
    },
    {
     "repo": "openbmb/MiniCPM5-2B",
     "commit": "42433e9",
     "name": "MiniCPM5-2B",
     "tag": "reasoning off · chat",
     "model_key": "minicpm5-2b",
     "maker": "OpenBMB",
     "kind": "chat",
     "params_total": 2.52,
     "params_active": 2.52,
     "quantization": "BF16",
     "score_0_100": -16.80640243902439,
     "raw": {
      "metric": "accuracy",
      "value": 0.06554878048780488,
      "questions": 656,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1299"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision openbmb/MiniCPM5-2B@42433e9), one question per request, plain answer line CEVAP: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking off; output cap 32,768"
    }
   ]
  },
  {
   "bench": "turkish-grammar",
   "title": "Turkish Grammar",
   "url": "https://zekiworks.github.io/analysis/decision-models/",
   "updated": "2026-11-16T20:50:26Z",
   "score": {
    "zero": "chance: 1 of 5 options right on every question (accuracy 20%)",
    "hundred": "every one of the 1,542 Turkish grammar questions (units 7 to 20) right (accuracy 100%)"
   },
   "models": [
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Carbon",
     "tag": "high · chat",
     "model_key": "gemini-4-carbon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 98.54085603112841,
     "raw": {
      "metric": "accuracy",
      "value": 0.9883268482490273,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.003112840466917862,
     "improvement_points": null,
     "runs": [
      "text:1108"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 3.8 Flash",
     "tag": "high · chat",
     "model_key": "gemini-3.8-flash",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 98.13553826199741,
     "raw": {
      "metric": "accuracy",
      "value": 0.9850843060959793,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.009889753566798921,
     "improvement_points": null,
     "runs": [
      "text:140",
      "text:123",
      "text:188"
     ],
     "notes": "Gemini API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Hydrogen",
     "tag": "high · chat",
     "model_key": "gemini-4-hydrogen",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 98.05447470817121,
     "raw": {
      "metric": "accuracy",
      "value": 0.9844357976653697,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.008767833981832101,
     "improvement_points": null,
     "runs": [
      "text:1114"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.2",
     "tag": "max · chat",
     "model_key": "claude-fable-5.2",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.89234760051879,
     "raw": {
      "metric": "accuracy",
      "value": 0.9831387808041504,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.07247081712061788,
     "improvement_points": null,
     "runs": [
      "text:980"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Astra",
     "tag": "high · chat",
     "model_key": "gpt-6-astra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.89234760051879,
     "raw": {
      "metric": "accuracy",
      "value": 0.9831387808041504,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.006554474708178415,
     "improvement_points": null,
     "runs": [
      "text:277",
      "text:610",
      "text:48",
      "text:191"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Sol",
     "tag": "high · chat",
     "model_key": "gpt-6.1-sol",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.89234760051879,
     "raw": {
      "metric": "accuracy",
      "value": 0.9831387808041504,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.00634241245136961,
     "improvement_points": null,
     "runs": [
      "text:276",
      "text:124",
      "text:190"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.1",
     "tag": "high · chat",
     "model_key": "claude-fable-5.1",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.4604409857328,
     "raw": {
      "metric": "accuracy",
      "value": 0.9636835278858625,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.12450064850843134,
     "improvement_points": null,
     "runs": [
      "text:310"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Argon",
     "tag": "low · chat",
     "model_key": "gemini-4-argon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.13618677042801,
     "raw": {
      "metric": "accuracy",
      "value": 0.9610894941634242,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.02749027237354718,
     "improvement_points": null,
     "runs": [
      "text:557"
     ],
     "notes": "Gemini API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Opus 5.5",
     "tag": "high · chat",
     "model_key": "claude-opus-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 93.43385214007782,
     "raw": {
      "metric": "accuracy",
      "value": 0.9474708171206225,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.14910505836575794,
     "improvement_points": null,
     "runs": [
      "text:278",
      "text:47",
      "text:192"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Terra",
     "tag": "high · chat",
     "model_key": "gpt-6.1-terra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 91.81258106355384,
     "raw": {
      "metric": "accuracy",
      "value": 0.9345006485084306,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.0353566796368462,
     "improvement_points": null,
     "runs": [
      "text:275",
      "text:274"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Grok 4.7",
     "tag": "high · chat",
     "model_key": "grok-4.7",
     "maker": "xAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 89.70492866407261,
     "raw": {
      "metric": "accuracy",
      "value": 0.917639429312581,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.12440337224383956,
     "improvement_points": null,
     "runs": [
      "text:311"
     ],
     "notes": "xAI Responses API, SuperGrok subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · first release",
     "model_key": "deepseek-v4.1-flash-first-release",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": 552.0,
     "params_active": 16.0,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 89.29961089494162,
     "raw": {
      "metric": "accuracy",
      "value": 0.914396887159533,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.08558556570328835,
     "improvement_points": null,
     "runs": [
      "text:121",
      "text:120"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; first-release weights (revision fb2764a), not the 1017 update"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · 1017",
     "model_key": "deepseek-v4.1-flash-1017",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 88.48897535667962,
     "raw": {
      "metric": "accuracy",
      "value": 0.9079118028534371,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.07282882248850497,
     "improvement_points": null,
     "runs": [
      "text:585",
      "text:558"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; 1017 weights (revision 1bae0e1)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen4-Max",
     "tag": "high · chat",
     "model_key": "qwen4-max",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 88.08365758754863,
     "raw": {
      "metric": "accuracy",
      "value": 0.9046692607003891,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.08305298013245119,
     "improvement_points": null,
     "runs": [
      "text:1336"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint (the maker's), fallbacks off, reasoning effort high, output cap 32,768 tokens, reasoning included; the answer schema"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Sonnet 5.5",
     "tag": "high · chat",
     "model_key": "claude-sonnet-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 81.19325551232166,
     "raw": {
      "metric": "accuracy",
      "value": 0.8495460440985733,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1692153047989607,
     "improvement_points": null,
     "runs": [
      "text:55",
      "text:193",
      "text:54"
     ],
     "notes": "Claude Code CLI, Claude subscription"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Muse Ember",
     "tag": "max · chat",
     "model_key": "muse-ember",
     "maker": "Meta",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 79.57198443579766,
     "raw": {
      "metric": "accuracy",
      "value": 0.8365758754863813,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06308567603748529,
     "improvement_points": null,
     "runs": [
      "text:939"
     ],
     "notes": "OpenRouter chat API, Meta's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning max"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.5-397B-A17B",
     "tag": "high · chat",
     "model_key": "qwen-3.5-397b-a17b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 397.0,
     "params_active": 17.0,
     "quantization": null,
     "score_0_100": 73.97859922178986,
     "raw": {
      "metric": "accuracy",
      "value": 0.791828793774319,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.09082360570687611,
     "improvement_points": null,
     "runs": [
      "text:870"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Kimi K3.1 Pro",
     "tag": "low · chat",
     "model_key": "kimi-k3.1-pro",
     "maker": "Moonshot",
     "kind": "chat",
     "params_total": 2800.0,
     "params_active": 104.0,
     "quantization": null,
     "score_0_100": 72.60051880674447,
     "raw": {
      "metric": "accuracy",
      "value": 0.7808041504539559,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.16202983138781726,
     "improvement_points": null,
     "runs": [
      "text:320",
      "text:315"
     ],
     "notes": "OpenRouter chat API, Moonshot AI's endpoint (mxfp4), fallbacks off"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen1.5",
     "commit": "8e56aa1b236a6dd1064dcfd22ba1e5e360e4c140",
     "name": "KnowLine-31B-Gen1.5",
     "tag": "",
     "model_key": "knowline-31b-gen1.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 71.38456549935148,
     "raw": {
      "metric": "accuracy",
      "value": 0.771076523994812,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.03856193907697914,
     "improvement_points": null,
     "runs": [
      "text:957"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "Poindexter-Labs/Copernicus-31B",
     "commit": "be71525",
     "name": "Copernicus-31B",
     "tag": "",
     "model_key": "copernicus-31b",
     "maker": "Poindexter Labs",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 71.22243839169909,
     "raw": {
      "metric": "accuracy",
      "value": 0.7697795071335928,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.10459595044202305,
     "improvement_points": null,
     "runs": [
      "text:1165"
     ],
     "notes": "Copernicus server (the card's copernicus_server.py, Transformers), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Inkling-Small",
     "tag": "high · chat · Jul 2026",
     "model_key": "inkling-small",
     "maker": "Thinking Machines",
     "kind": "chat",
     "params_total": 265.96,
     "params_active": 12.0,
     "quantization": null,
     "score_0_100": 71.14137483787289,
     "raw": {
      "metric": "accuracy",
      "value": 0.7691309987029832,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.09922178988326955,
     "improvement_points": null,
     "runs": [
      "text:866"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included; no response format (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen2",
     "commit": "fd99a9e",
     "name": "KnowLine-31B-Gen2",
     "tag": "",
     "model_key": "knowline-31b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 70.00648508430609,
     "raw": {
      "metric": "accuracy",
      "value": 0.7600518806744487,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.056990138293546336,
     "improvement_points": null,
     "runs": [
      "text:1314"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "turion-ai/turion-jev-31b",
     "commit": "866c0fc6e040cba23fcc72113a80e5c99d5ebab1",
     "name": "Turion Jev 31B",
     "tag": "",
     "model_key": "turion-jev-31b",
     "maker": "Turion.AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 at load",
     "score_0_100": 68.46627756160831,
     "raw": {
      "metric": "accuracy",
      "value": 0.7477302204928664,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.12005543904020738,
     "improvement_points": null,
     "runs": [
      "text:1273"
     ],
     "notes": "Turion Jev server (turion_jev.serve on vLLM, the maker's release code), self-hosted, FP8 at load"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3-Flash",
     "tag": "reasoning on · chat",
     "model_key": "glm-5.3-flash",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 320.0,
     "params_active": 18.0,
     "quantization": "EXL3 4-bit experts, FP8 dense layers and KV cache",
     "score_0_100": 67.89883268482491,
     "raw": {
      "metric": "accuracy",
      "value": 0.7431906614785992,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.043826025459691616,
     "improvement_points": null,
     "runs": [
      "text:160",
      "text:159"
     ],
     "notes": "TensorFold, self-hosted, EXL3 4-bit experts"
    },
    {
     "repo": "decisionmodels/REAX-1.3",
     "commit": "fa4bf7e",
     "name": "REAX 1.3",
     "tag": "",
     "model_key": "reax-1.3",
     "maker": "REAX",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 64.7373540856031,
     "raw": {
      "metric": "accuracy",
      "value": 0.7178988326848249,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.050560403571647254,
     "improvement_points": null,
     "runs": [
      "text:827"
     ],
     "notes": "REAX Engine server (reax13 mode, the maker's release code), self-hosted, FP8"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini Decision 1",
     "tag": "20261111",
     "model_key": "gemini-decision-1",
     "maker": "Google",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 64.1699092088197,
     "raw": {
      "metric": "accuracy",
      "value": 0.7133592736705577,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.029234586461398322,
     "improvement_points": null,
     "runs": [
      "text:1021"
     ],
     "notes": "OpenRouter decisions API, Google AI Studio endpoint"
    },
    {
     "repo": "MachinePulseAI/karpo-decision-31B",
     "commit": "d69cdd5",
     "name": "Karpo Decision 31B",
     "tag": "",
     "model_key": "karpo-decision-31b",
     "maker": "MachinePulse",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 63.92671854734111,
     "raw": {
      "metric": "accuracy",
      "value": 0.711413748378729,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.03372972747051323,
     "improvement_points": null,
     "runs": [
      "text:1156"
     ],
     "notes": "karpo-decision server (the System One server of the card's karpo-decision package, machinepulse-ai/karpo-decision at ac40d39), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1-ultra",
     "tag": "",
     "model_key": "liquid-d1-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 63.35927367055771,
     "raw": {
      "metric": "accuracy",
      "value": 0.7068741893644618,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.042313994668471085,
     "improvement_points": null,
     "runs": [
      "text:199"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3",
     "tag": "low · chat",
     "model_key": "glm-5.3",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 753.33,
     "params_active": 753.33,
     "quantization": null,
     "score_0_100": 61.17055771725032,
     "raw": {
      "metric": "accuracy",
      "value": 0.6893644617380026,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.054636835278861325,
     "improvement_points": null,
     "runs": [
      "text:314"
     ],
     "notes": "OpenRouter chat API, Fireworks endpoint, fallbacks off"
    },
    {
     "repo": "Qwen/Qwen3.8-27B",
     "commit": "1d4bf0f",
     "name": "Qwen3.8-27B",
     "tag": "low · chat",
     "model_key": "qwen3.8-27b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 27.78,
     "params_active": 27.78,
     "quantization": "BF16",
     "score_0_100": 60.52204928664072,
     "raw": {
      "metric": "accuracy",
      "value": 0.6841763942931258,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.2941992418036467,
     "improvement_points": null,
     "runs": [
      "text:318",
      "text:50",
      "text:53",
      "text:83"
     ],
     "notes": "vLLM, self-hosted"
    },
    {
     "repo": "logitshq/Logits-Sigma-1-Duo",
     "commit": "a0c1b6e",
     "name": "Logits Sigma-1 Duo",
     "tag": "",
     "model_key": "logits-sigma-1-duo",
     "maker": "Logits",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 59.79247730220493,
     "raw": {
      "metric": "accuracy",
      "value": 0.6783398184176395,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.059110714035682306,
     "improvement_points": null,
     "runs": [
      "text:1221"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "sthanika-ai/Sieve-31B-v3",
     "commit": "8fcee858f7f20a50bfa3c8f984c22adb3cbb46a2",
     "name": "Sieve 31B v3",
     "tag": "",
     "model_key": "sieve-31b-v3",
     "maker": "Sthānika AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 58.33333333333333,
     "raw": {
      "metric": "accuracy",
      "value": 0.6666666666666666,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.033395719844357886,
     "improvement_points": null,
     "runs": [
      "text:1241"
     ],
     "notes": "Sieve server (sieve-serve, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiMo-V2.6-Flash",
     "tag": "high · chat",
     "model_key": "mimo-v2.6-flash",
     "maker": "Xiaomi",
     "kind": "chat",
     "params_total": 309.0,
     "params_active": 15.0,
     "quantization": null,
     "score_0_100": 57.44163424124513,
     "raw": {
      "metric": "accuracy",
      "value": 0.6595330739299611,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.10338066260987085,
     "improvement_points": null,
     "runs": [
      "text:877"
     ],
     "notes": "OpenRouter chat API, Xiaomi's own endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included"
    },
    {
     "repo": "google/gemma-4-31B-it",
     "commit": "842da37",
     "name": "Gemma 4 31B",
     "tag": "reasoning off · chat",
     "model_key": "gemma-4-31b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 weights and KV cache",
     "score_0_100": 56.95525291828793,
     "raw": {
      "metric": "accuracy",
      "value": 0.6556420233463035,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.30996504491707116,
     "improvement_points": null,
     "runs": [
      "text:119",
      "text:189"
     ],
     "notes": "vLLM, self-hosted, FP8"
    },
    {
     "repo": "fastino/GLiDE-27B",
     "commit": "4d32743",
     "name": "GLiDE 27B",
     "tag": "",
     "model_key": "glide-27b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 53.38845654993515,
     "raw": {
      "metric": "accuracy",
      "value": 0.6271076523994812,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.026675864550280738,
     "improvement_points": null,
     "runs": [
      "text:899"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.2",
     "tag": "API · with web",
     "model_key": "perplexity-decider-27b-v1.2-with-web",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 50.22697795071335,
     "raw": {
      "metric": "accuracy",
      "value": 0.6018158236057068,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.04854037190444724,
     "improvement_points": null,
     "runs": [
      "text:640"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "low · chat",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 50.22697795071335,
     "raw": {
      "metric": "accuracy",
      "value": 0.6018158236057068,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.3274448767834053,
     "improvement_points": null,
     "runs": [
      "text:133"
     ],
     "notes": "Codex CLI, ChatGPT subscription"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.3",
     "commit": "9c2ebb2",
     "name": "Perplexity Decider 27B v1.3",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 49.57846952010376,
     "raw": {
      "metric": "accuracy",
      "value": 0.5966277561608301,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06077237712993911,
     "improvement_points": null,
     "runs": [
      "text:1134"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "fastino/GLiDE",
     "commit": "a7e4589",
     "name": "GLiDE",
     "tag": "v1.2",
     "model_key": "fastino-glide-v1.2",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 49.33527885862516,
     "raw": {
      "metric": "accuracy",
      "value": 0.5946822308690013,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.030490740858430832,
     "improvement_points": null,
     "runs": [
      "text:248"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiniMax-M3.1-Flash",
     "tag": "high · chat",
     "model_key": "minimax-m3.1-flash",
     "maker": "MiniMax",
     "kind": "chat",
     "params_total": 428.0,
     "params_active": 23.0,
     "quantization": null,
     "score_0_100": 49.33527885862516,
     "raw": {
      "metric": "accuracy",
      "value": 0.5946822308690013,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.26133116883117197,
     "improvement_points": null,
     "runs": [
      "text:862"
     ],
     "notes": "OpenRouter chat API, MiniMax's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning included; JSON mode without the answer schema (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLiDE",
     "tag": "API",
     "model_key": "fastino-glide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 49.25421530479896,
     "raw": {
      "metric": "accuracy",
      "value": 0.5940337224383917,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.04076048164813012,
     "improvement_points": null,
     "runs": [
      "text:129"
     ],
     "notes": "Fastino System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2-ultra",
     "tag": "",
     "model_key": "liquid-d1.2-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 48.605706874189366,
     "raw": {
      "metric": "accuracy",
      "value": 0.5888456549935149,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.03145185745941194,
     "improvement_points": null,
     "runs": [
      "text:1347"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": "vllm-sr/d3",
     "commit": "20f37b9",
     "name": "d3",
     "tag": "",
     "model_key": "vllm-sr-d3",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 48.362516212710766,
     "raw": {
      "metric": "accuracy",
      "value": 0.5869001297016861,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.024076136755969275,
     "improvement_points": null,
     "runs": [
      "text:259"
     ],
     "notes": "vLLM System One endpoint (nightly), self-hosted, BF16"
    },
    {
     "repo": "vllm-sr/d3-ultra",
     "commit": "52ba9e7",
     "name": "d3-ultra",
     "tag": "",
     "model_key": "vllm-sr-d3-ultra",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 48.03826199740597,
     "raw": {
      "metric": "accuracy",
      "value": 0.5843060959792478,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.16264256879080513,
     "improvement_points": null,
     "runs": [
      "text:269"
     ],
     "notes": "d3 package server (transformers, the release code), self-hosted, BF16"
    },
    {
     "repo": "nace-ai/drex-max-v1",
     "commit": "cac65e1",
     "name": "Drex Max",
     "tag": "",
     "model_key": "drex-max-v1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 48.03826199740597,
     "raw": {
      "metric": "accuracy",
      "value": 0.5843060959792478,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.07970248845472344,
     "improvement_points": null,
     "runs": [
      "text:322"
     ],
     "notes": "Drex Max server (Nace's loader, drex_max), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex Max v1.1",
     "tag": "",
     "model_key": "drex-max-1.1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 47.95719844357977,
     "raw": {
      "metric": "accuracy",
      "value": 0.5836575875486382,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.027224315746671758,
     "improvement_points": null,
     "runs": [
      "text:326"
     ],
     "notes": "Nace.AI System One API"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.2",
     "commit": "225c913",
     "name": "Perplexity Decider 27B v1.2",
     "tag": "DGX Spark",
     "model_key": "perplexity-decider-27b-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 47.06549935149156,
     "raw": {
      "metric": "accuracy",
      "value": 0.5765239948119325,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06172293290366029,
     "improvement_points": null,
     "runs": [
      "text:217",
      "text:633"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 46.82230869001296,
     "raw": {
      "metric": "accuracy",
      "value": 0.5745784695201037,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.05461854631580725,
     "improvement_points": null,
     "runs": [
      "text:145"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": "baselabs/mammoth-1",
     "commit": "61cec44",
     "name": "Mammoth-1",
     "tag": "",
     "model_key": "mammoth-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 46.49805447470816,
     "raw": {
      "metric": "accuracy",
      "value": 0.5719844357976653,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.07895643579766512,
     "improvement_points": null,
     "runs": [
      "text:374"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap Preview",
     "tag": "",
     "model_key": "subq-snap-preview",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 44.63359273670557,
     "raw": {
      "metric": "accuracy",
      "value": 0.5570687418936446,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06016216167138863,
     "improvement_points": null,
     "runs": [
      "text:177"
     ],
     "notes": "SubQ Decisions API (public preview)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "Decisions API",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 42.850194552529175,
     "raw": {
      "metric": "accuracy",
      "value": 0.5428015564202334,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.0303638726445742,
     "improvement_points": null,
     "runs": [
      "text:143"
     ],
     "notes": "OpenAI Decisions API (public beta)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap 1.1",
     "tag": "",
     "model_key": "subq-snap-1.1",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 42.52594033722438,
     "raw": {
      "metric": "accuracy",
      "value": 0.540207522697795,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.14546577686542977,
     "improvement_points": null,
     "runs": [
      "text:346"
     ],
     "notes": "SubQ Decisions API (public preview)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.8 Flash",
     "tag": "no reasoning · chat",
     "model_key": "qwen3.8-flash",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 40.985732814526585,
     "raw": {
      "metric": "accuracy",
      "value": 0.5278858625162127,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.385136186770434,
     "improvement_points": null,
     "runs": [
      "text:985"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint, fallbacks off, reasoning off"
    },
    {
     "repo": "PelaAI/KnowLine-12B-Gen1",
     "commit": "f26fbc2",
     "name": "KnowLine-12B-Gen1",
     "tag": "",
     "model_key": "knowline-12b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 40.58041504539559,
     "raw": {
      "metric": "accuracy",
      "value": 0.5246433203631647,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.043251979112706694,
     "improvement_points": null,
     "runs": [
      "text:1309"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.1",
     "tag": "",
     "model_key": "liquid-d1.1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 39.607652399481196,
     "raw": {
      "metric": "accuracy",
      "value": 0.5168612191958496,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.022286633304368423,
     "improvement_points": null,
     "runs": [
      "text:254"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2",
     "tag": "",
     "model_key": "liquid-d1.2",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 39.283398184176384,
     "raw": {
      "metric": "accuracy",
      "value": 0.5142671854734111,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.02507946857214791,
     "improvement_points": null,
     "runs": [
      "text:1346"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning on · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 37.90531776913099,
     "raw": {
      "metric": "accuracy",
      "value": 0.503242542153048,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1149"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking on; output cap 32,768"
    },
    {
     "repo": null,
     "commit": null,
     "name": "celeris-1-decision",
     "tag": "",
     "model_key": "celeris-1-decision",
     "maker": "Celeris",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 37.418936446173795,
     "raw": {
      "metric": "accuracy",
      "value": 0.4993514915693904,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.053249027237353946,
     "improvement_points": null,
     "runs": [
      "text:150"
     ],
     "notes": "Celeris System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 36.8514915693904,
     "raw": {
      "metric": "accuracy",
      "value": 0.49481193255512324,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.050289363215129355,
     "improvement_points": null,
     "runs": [
      "text:44"
     ],
     "notes": "Perplexity Decisions API"
    },
    {
     "repo": "google/gemma-4-E4B-it",
     "commit": "4ecc2ba",
     "name": "Gemma 4 E4B",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-e4b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 8.0,
     "params_active": 8.0,
     "quantization": "BF16",
     "score_0_100": 36.284046692607,
     "raw": {
      "metric": "accuracy",
      "value": 0.490272373540856,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1040"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E4B-it@4ecc2ba), one question per request, plain answer line CEVAP: X (no answer schema); sampling temperature 0.0; output cap 8,192"
    },
    {
     "repo": "OmniJev/OneJev-27B",
     "commit": "a5e5f55",
     "name": "OneJev-27B",
     "tag": "",
     "model_key": "onejev-27b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 35.878728923476,
     "raw": {
      "metric": "accuracy",
      "value": 0.48702983138780803,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.0371355376134889,
     "improvement_points": null,
     "runs": [
      "text:264"
     ],
     "notes": "OneJev server (qev, OmniJev's release code), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.14",
     "tag": "",
     "model_key": "jev-1.14.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 35.2302204928664,
     "raw": {
      "metric": "accuracy",
      "value": 0.48184176394293127,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.039993514915693806,
     "improvement_points": null,
     "runs": [
      "text:194"
     ],
     "notes": "TypeSafe System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Mercury Decide",
     "tag": "",
     "model_key": "inception-mercury-decide",
     "maker": "Inception",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 33.93320363164721,
     "raw": {
      "metric": "accuracy",
      "value": 0.4714656290531777,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1895821213231088,
     "improvement_points": null,
     "runs": [
      "text:204"
     ],
     "notes": "OpenRouter System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.13",
     "tag": "",
     "model_key": "jev-1.13.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 32.960440985732816,
     "raw": {
      "metric": "accuracy",
      "value": 0.46368352788586253,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06547989623865107,
     "improvement_points": null,
     "runs": [
      "text:49"
     ],
     "notes": "TypeSafe System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1",
     "tag": "",
     "model_key": "liquid-d1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 32.555123216601814,
     "raw": {
      "metric": "accuracy",
      "value": 0.4604409857328145,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.13508827516330688,
     "improvement_points": null,
     "runs": [
      "text:52"
     ],
     "notes": "Liquid AI decisions API"
    },
    {
     "repo": "tattrongvu/Jeffan-9B-1211",
     "commit": "862d1a7",
     "name": "Jeffan-9B-1211",
     "tag": "",
     "model_key": "jeffan-9b-1211",
     "maker": "tattrongvu",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 32.47405966277561,
     "raw": {
      "metric": "accuracy",
      "value": 0.4597924773022049,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.030358104375877916,
     "improvement_points": null,
     "runs": [
      "text:1192"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "Cloudflare/clef",
     "commit": "2f3de3d",
     "name": "Clef",
     "tag": "",
     "model_key": "clef",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 27.48,
     "params_active": 27.48,
     "quantization": "BF16",
     "score_0_100": 31.825551232166017,
     "raw": {
      "metric": "accuracy",
      "value": 0.45460440985732814,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1511928015564202,
     "improvement_points": null,
     "runs": [
      "text:125"
     ],
     "notes": "Clef server, self-hosted"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex 2.0",
     "tag": "",
     "model_key": "drex-2.0",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 35.11,
     "params_active": 3.0,
     "quantization": null,
     "score_0_100": 31.014915693904022,
     "raw": {
      "metric": "accuracy",
      "value": 0.4481193255512322,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.01956452658884572,
     "improvement_points": null,
     "runs": [
      "text:334"
     ],
     "notes": "Nace.AI System One API"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Haiku 5.5",
     "tag": "max · chat",
     "model_key": "claude-haiku-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 30.447470817120614,
     "raw": {
      "metric": "accuracy",
      "value": 0.44357976653696496,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.16387341772151878,
     "improvement_points": null,
     "runs": [
      "text:948"
     ],
     "notes": "OpenRouter chat API, Anthropic's own endpoint, fallbacks off, output cap 128,000 tokens, reasoning max"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Cygnet",
     "tag": "",
     "model_key": "cygnet",
     "maker": "Blockbrain Labs",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 30.204280155642017,
     "raw": {
      "metric": "accuracy",
      "value": 0.44163424124513617,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.189059729699381,
     "improvement_points": null,
     "runs": [
      "text:127"
     ],
     "notes": "Cygnet decision server on vLLM, self-hosted"
    },
    {
     "repo": "EldanRing/Winnow-12B",
     "commit": "b2b1421",
     "name": "Winnow-12B",
     "tag": "",
     "model_key": "winnow-12b",
     "maker": "EldanRing",
     "kind": "decision",
     "params_total": 11.91,
     "params_active": 11.91,
     "quantization": "Q8_0",
     "score_0_100": 30.204280155642017,
     "raw": {
      "metric": "accuracy",
      "value": 0.44163424124513617,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.24128724281679428,
     "improvement_points": null,
     "runs": [
      "text:128"
     ],
     "notes": "Winnow server (llama.cpp), self-hosted"
    },
    {
     "repo": "ZefanCai/Open-Jev-27B-v1.1",
     "commit": "28cf730",
     "name": "Open-Jev-27B-v1.1",
     "tag": "",
     "model_key": "open-jev-27b-v1.1",
     "maker": "Zefan Cai",
     "kind": "decision",
     "params_total": 27.8,
     "params_active": 27.8,
     "quantization": "BF16",
     "score_0_100": 30.123216601815816,
     "raw": {
      "metric": "accuracy",
      "value": 0.4409857328145266,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.16411523543511222,
     "improvement_points": null,
     "runs": [
      "text:45"
     ],
     "notes": "Open-Jev server, self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen1",
     "commit": "325378d",
     "name": "KnowLine-9B-Gen1",
     "tag": "",
     "model_key": "knowline-9b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 29.71789883268482,
     "raw": {
      "metric": "accuracy",
      "value": 0.4377431906614786,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06763983986109458,
     "improvement_points": null,
     "runs": [
      "text:1174"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "apertixai/cerno-9b-v1.2",
     "commit": "f02a632fbe3a4698cd1ff5c0b929ea240e2b8eda",
     "name": "Cerno 9B v1.2",
     "tag": "",
     "model_key": "cerno-9b-v1.2",
     "maker": "Apertix AI",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 29.63683527885862,
     "raw": {
      "metric": "accuracy",
      "value": 0.437094682230869,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.142308560311284,
     "improvement_points": null,
     "runs": [
      "text:1201"
     ],
     "notes": "Cerno server (a thin System One server around the card's CernoEngine, transformers: logic's for its English runs, compactor's di_serve.py for its Turkish runs), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen2",
     "commit": "8ec8eb6",
     "name": "KnowLine-9B-Gen2",
     "tag": "",
     "model_key": "knowline-9b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 27.610246433203628,
     "raw": {
      "metric": "accuracy",
      "value": 0.42088197146562906,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.051156884569114935,
     "improvement_points": null,
     "runs": [
      "text:1306"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted"
    },
    {
     "repo": "newsbang/adjudex-9b-v5",
     "commit": "ffdcb82",
     "name": "Adjudex 9B v5",
     "tag": "",
     "model_key": "adjudex-9b-v5",
     "maker": "newsbang",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 26.71854734111543,
     "raw": {
      "metric": "accuracy",
      "value": 0.41374837872892345,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.12177337512381844,
     "improvement_points": null,
     "runs": [
      "text:1183"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "baselabs/elephant-1",
     "commit": "b3f8a93",
     "name": "Elephant-1",
     "tag": "",
     "model_key": "elephant-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 9.41,
     "params_active": 9.41,
     "quantization": "BF16",
     "score_0_100": 25.50259403372244,
     "raw": {
      "metric": "accuracy",
      "value": 0.4040207522697795,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.15412009857328138,
     "improvement_points": null,
     "runs": [
      "text:373"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16"
    },
    {
     "repo": "fastino/GLiDE-9B",
     "commit": "ab54df5",
     "name": "GLiDE 9B",
     "tag": "",
     "model_key": "glide-9b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 7.94,
     "params_active": 7.94,
     "quantization": "BF16",
     "score_0_100": 25.09727626459144,
     "raw": {
      "metric": "accuracy",
      "value": 0.40077821011673154,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.11989879551751426,
     "improvement_points": null,
     "runs": [
      "text:907"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "Patrick-18/Dexy-Decision-9B-V2",
     "commit": "4f434e2",
     "name": "Dexy-Decision-9B-V2",
     "tag": "",
     "model_key": "dexy-decision-9b-v2",
     "maker": "Patrick-18",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 24.610894941634236,
     "raw": {
      "metric": "accuracy",
      "value": 0.3968871595330739,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1272381252256,
     "improvement_points": null,
     "runs": [
      "text:1255"
     ],
     "notes": "Dexy server (the card's Transformers d3_server.py, in its requirements.txt pins: torch 2.14.0, transformers 5.19.0), self-hosted, BF16"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-9B-Instant",
     "commit": "18a17f734966b213b2f00fa9c915e8b9af2a9b20",
     "name": "StartLux-Decision-v1.1-9B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-9b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 24.36770428015564,
     "raw": {
      "metric": "accuracy",
      "value": 0.3949416342412451,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.11323279756488182,
     "improvement_points": null,
     "runs": [
      "text:1290"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.3",
     "commit": "d06fb3f",
     "name": "Perplexity Decider Flash v1.3",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 24.04345006485084,
     "raw": {
      "metric": "accuracy",
      "value": 0.39234760051880674,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.15315801264982284,
     "improvement_points": null,
     "runs": [
      "text:1123"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "scar-ai/vera-medium-omni",
     "commit": "1f23a6c",
     "name": "vera-medium-omni",
     "tag": "",
     "model_key": "vera-medium-omni",
     "maker": "scar-ai",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 22.341115434500644,
     "raw": {
      "metric": "accuracy",
      "value": 0.3787289234760052,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.07954954604409842,
     "improvement_points": null,
     "runs": [
      "text:1264"
     ],
     "notes": "vera server (the card's own vera-serve from vera-s1 0.1.8, PyTorch engine), self-hosted, BF16"
    },
    {
     "repo": "PelaAI/KnowLine-4B-Gen5.5",
     "commit": "779944d2cc8ffa19cef386aee83c83bcb967ade2",
     "name": "KnowLine 4B Gen5.5",
     "tag": "",
     "model_key": "knowline-4b-gen5.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "FP8 at load",
     "score_0_100": 21.61154345006485,
     "raw": {
      "metric": "accuracy",
      "value": 0.3728923476005188,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.16567138646668936,
     "improvement_points": null,
     "runs": [
      "text:965"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, FP8 at load"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.2",
     "commit": "6d771ad",
     "name": "Perplexity Decider Flash v1.2",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 21.44941634241245,
     "raw": {
      "metric": "accuracy",
      "value": 0.3715953307392996,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.10969746921858275,
     "improvement_points": null,
     "runs": [
      "text:223"
     ],
     "notes": "Perplexity Decider server, self-hosted"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-4B-Instant",
     "commit": "416088f77cfce09633906a3675c7cd2e5dabcf10",
     "name": "StartLux-Decision-v1.1-4B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-4b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 20.88197146562905,
     "raw": {
      "metric": "accuracy",
      "value": 0.3670557717250324,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.10274788153319536,
     "improvement_points": null,
     "runs": [
      "text:1236"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 20.88197146562905,
     "raw": {
      "metric": "accuracy",
      "value": 0.3670557717250324,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1146"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768"
    },
    {
     "repo": "victorbennett/Verdin-4B-v1.1",
     "commit": "45830d2",
     "name": "Verdin-4B-v1.1",
     "tag": "",
     "model_key": "verdin-4b-v1.1",
     "maker": "victorbennett",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 20.47665369649805,
     "raw": {
      "metric": "accuracy",
      "value": 0.3638132295719844,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.12760401020475481,
     "improvement_points": null,
     "runs": [
      "text:1244"
     ],
     "notes": "Verdin server (the card's own python -m verdin.server, in the package versions its card lists), self-hosted, BF16"
    },
    {
     "repo": "quincyqiang/chinese-decision-pointer-4b",
     "commit": "889f23e",
     "name": "Chinese Decision Pointer 4B",
     "tag": "",
     "model_key": "chinese-decision-pointer-4b",
     "maker": "quincyqiang",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 20.47665369649805,
     "raw": {
      "metric": "accuracy",
      "value": 0.3638132295719844,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.153621115989297,
     "improvement_points": null,
     "runs": [
      "text:1206"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16"
    },
    {
     "repo": "Valen-Team/Valen-Decision-1.5-9B",
     "commit": "5a51211",
     "name": "Valen-Decision-1.5-9B",
     "tag": "",
     "model_key": "valen-decision-1.5-9b",
     "maker": "Valen-Team",
     "kind": "decision",
     "params_total": 8.4,
     "params_active": 8.4,
     "quantization": "BF16",
     "score_0_100": 18.774319066147854,
     "raw": {
      "metric": "accuracy",
      "value": 0.35019455252918286,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06866481370102538,
     "improvement_points": null,
     "runs": [
      "text:1275"
     ],
     "notes": "Valen server (the card's valen-vllm plugin, Valen-Team/Valen at 773d3a6, in vllm/vllm-openai v0.33.0), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Microsoft-Decision-1",
     "tag": "",
     "model_key": "microsoft-decision-1",
     "maker": "Microsoft",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 18.693255512321656,
     "raw": {
      "metric": "accuracy",
      "value": 0.34954604409857326,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.24428434172536403,
     "improvement_points": null,
     "runs": [
      "text:205"
     ],
     "notes": "OpenRouter System One API"
    },
    {
     "repo": "Qwen/Qwen3-14B",
     "commit": "40c0698",
     "name": "Qwen3-14B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3-14b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 18.125810635538258,
     "raw": {
      "metric": "accuracy",
      "value": 0.3450064850843061,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.485785092279197,
     "improvement_points": null,
     "runs": [
      "text:131"
     ],
     "notes": "vLLM, self-hosted, BF16"
    },
    {
     "repo": "Cloudflare/clef-flash",
     "commit": "fde727a",
     "name": "Clef-Flash",
     "tag": "",
     "model_key": "clef-flash",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 9.53,
     "params_active": 9.53,
     "quantization": "BF16",
     "score_0_100": 17.63942931258106,
     "raw": {
      "metric": "accuracy",
      "value": 0.3411154345006485,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.20301802853437087,
     "improvement_points": null,
     "runs": [
      "text:139"
     ],
     "notes": "Clef server, self-hosted"
    },
    {
     "repo": "ecloudtech/Erk-14B",
     "commit": "8584b14",
     "name": "Erk-14B",
     "tag": "reasoning off · chat",
     "model_key": "erk-14b",
     "maker": "eCloud Technology",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 16.747730220492862,
     "raw": {
      "metric": "accuracy",
      "value": 0.3339818417639429,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1632142058757866,
     "improvement_points": null,
     "runs": [
      "text:132"
     ],
     "notes": "vLLM, self-hosted, BF16"
    },
    {
     "repo": "wayfind/metask-jev-4b-policy-mix",
     "commit": "ea20fe8",
     "name": "Metask-Jev-4B",
     "tag": "",
     "model_key": "metask-jev-4b",
     "maker": "Metask Lab",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 14.234760051880672,
     "raw": {
      "metric": "accuracy",
      "value": 0.3138780804150454,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06542194194985428,
     "improvement_points": null,
     "runs": [
      "text:126"
     ],
     "notes": "Metask-Jev server, self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-2B-Gen1",
     "commit": "53245d470a4ce316e32fd0d18c24e142c697f40b",
     "name": "KnowLine-2B-Gen1",
     "tag": "",
     "model_key": "knowline-2b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 11.721789883268482,
     "raw": {
      "metric": "accuracy",
      "value": 0.29377431906614787,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.11350557340970796,
     "improvement_points": null,
     "runs": [
      "text:1079"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "LiquidAI/LFM2.5-2.6B",
     "commit": "7141b56",
     "name": "LFM2.5-2.6B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-2.6b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 2.7,
     "params_active": 2.7,
     "quantization": "BF16",
     "score_0_100": 11.235408560311285,
     "raw": {
      "metric": "accuracy",
      "value": 0.2898832684824903,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1037"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-2.6B@7141b56), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50; output cap 32,768"
    },
    {
     "repo": "fastino/GLiDE-2B",
     "commit": "00f122c",
     "name": "GLiDE 2B",
     "tag": "",
     "model_key": "glide-2b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 1.88,
     "params_active": 1.88,
     "quantization": "BF16",
     "score_0_100": 10.667963683527883,
     "raw": {
      "metric": "accuracy",
      "value": 0.2853437094682231,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.22548275064105452,
     "improvement_points": null,
     "runs": [
      "text:915"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning on · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 9.208819714656293,
     "raw": {
      "metric": "accuracy",
      "value": 0.27367055771725035,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1089"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking on; output cap 32,768"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-2B-Instant",
     "commit": "f1e6cd147dda82c9ffa3e8e725d5c1f175566fb1",
     "name": "StartLux-Decision-v1.1-2B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-2b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 7.911802853437094,
     "raw": {
      "metric": "accuracy",
      "value": 0.26329442282749677,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.09365771544180286,
     "improvement_points": null,
     "runs": [
      "text:977"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-0.8B-Instant",
     "commit": "6496ae641cc71b149ce9dad69bd97f2a44e7bf11",
     "name": "StartLux-Decision-v1.1-0.8B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-0.8b-instant",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 7.2632944228274985,
     "raw": {
      "metric": "accuracy",
      "value": 0.25810635538262,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.04647690514605797,
     "improvement_points": null,
     "runs": [
      "text:1015"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16"
    },
    {
     "repo": "OmniJev/OneJev1.6-0.8B",
     "commit": "bcd88a382f314b26cb7b1b7f369c1ed3502ee7a0",
     "name": "OneJev1.6-0.8B",
     "tag": "",
     "model_key": "onejev1.6-0.8b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 0.85,
     "params_active": 0.85,
     "quantization": "BF16",
     "score_0_100": 5.804150453955901,
     "raw": {
      "metric": "accuracy",
      "value": 0.24643320363164722,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.10343618871595343,
     "improvement_points": null,
     "runs": [
      "text:1029"
     ],
     "notes": "Flide server (flide 0.1.11, the card's server for OneJev 1.6), self-hosted, BF16"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning off · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 5.723086900129699,
     "raw": {
      "metric": "accuracy",
      "value": 0.2457846952010376,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1227"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking off; output cap 32,768"
    },
    {
     "repo": "LiquidAI/d1-2B",
     "commit": "10e7671",
     "name": "d1-2B",
     "tag": "",
     "model_key": "liquid-d1-2b",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": 1.92,
     "params_active": 1.92,
     "quantization": "BF16",
     "score_0_100": 5.6420233463035,
     "raw": {
      "metric": "accuracy",
      "value": 0.245136186770428,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.06678220316566075,
     "improvement_points": null,
     "runs": [
      "text:340"
     ],
     "notes": "d1 server (Liquid's model code in Transformers, the model card's call), self-hosted, BF16"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Decider 1",
     "tag": "",
     "model_key": "meragpt-decider-1",
     "maker": "meraGPT",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 4.912451361867705,
     "raw": {
      "metric": "accuracy",
      "value": 0.23929961089494164,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.15385077821011647,
     "improvement_points": null,
     "runs": [
      "text:243"
     ],
     "notes": "meraGPT System One API"
    },
    {
     "repo": "fastino/GLiDE-0.8B",
     "commit": "9eacdee",
     "name": "GLiDE 0.8B",
     "tag": "",
     "model_key": "glide-0.8b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.75,
     "params_active": 0.75,
     "quantization": "BF16",
     "score_0_100": 4.588197146562903,
     "raw": {
      "metric": "accuracy",
      "value": 0.23670557717250323,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.26070633214637023,
     "improvement_points": null,
     "runs": [
      "text:923"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "StrandsAgents/strands-decider-2B-hobson-v21",
     "commit": "2b52a62",
     "name": "Strands Decider 2B",
     "tag": "",
     "model_key": "strands-decider-2b",
     "maker": "Strands Agents",
     "kind": "decision",
     "params_total": 2.29,
     "params_active": 2.29,
     "quantization": "BF16",
     "score_0_100": 4.345006485084305,
     "raw": {
      "metric": "accuracy",
      "value": 0.23476005188067445,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.13162107652399468,
     "improvement_points": null,
     "runs": [
      "text:141"
     ],
     "notes": "Strands Decider server, self-hosted"
    },
    {
     "repo": "PelaAI/KnowLine-0.8B-Gen1",
     "commit": "6b27831f7b783c03278bd946d4d8f4f7a98a3798",
     "name": "KnowLine-0.8B-Gen1",
     "tag": "",
     "model_key": "knowline-0.8b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 2.96692607003891,
     "raw": {
      "metric": "accuracy",
      "value": 0.2237354085603113,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1078046042323395,
     "improvement_points": null,
     "runs": [
      "text:1005"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16"
    },
    {
     "repo": "knowledgator/gliclass-v4-edge",
     "commit": "885af1af8a07edf4ed34df83ca0d995ed8201283",
     "name": "GLiClass v4 Edge",
     "tag": "",
     "model_key": "gliclass-v4-edge",
     "maker": "Knowledgator",
     "kind": "decision",
     "params_total": 0.44,
     "params_active": 0.44,
     "quantization": "BF16",
     "score_0_100": 2.804798962386508,
     "raw": {
      "metric": "accuracy",
      "value": 0.22243839169909208,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.09974899224152792,
     "improvement_points": null,
     "runs": [
      "text:1046"
     ],
     "notes": "GLiClass server (our servers/gliclass_server.py: the card's gliclass pipeline and its System One mapping), self-hosted, BF16"
    },
    {
     "repo": "Qwen/Qwen3.5-2B",
     "commit": "15852e8",
     "name": "Qwen3.5-2B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-2b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 1.9941634241245134,
     "raw": {
      "metric": "accuracy",
      "value": 0.21595330739299612,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1218"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-2B@15852e8), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768"
    },
    {
     "repo": "fastino/GLiDE-0.4B",
     "commit": "6bda7ab",
     "name": "GLiDE 0.4B",
     "tag": "",
     "model_key": "glide-0.4b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.4,
     "params_active": 0.4,
     "quantization": "BF16",
     "score_0_100": 1.5888456549935128,
     "raw": {
      "metric": "accuracy",
      "value": 0.2127107652399481,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.0826740581414577,
     "improvement_points": null,
     "runs": [
      "text:931"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted"
    },
    {
     "repo": "fastino/GLiNER2.5-multi-Decide",
     "commit": "a35a0cd",
     "name": "GLiNER2.5-multi-Decide",
     "tag": "",
     "model_key": "gliner2.5-multi-decide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": -0.19455252918288216,
     "raw": {
      "metric": "accuracy",
      "value": 0.19844357976653695,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.22690143405270174,
     "improvement_points": null,
     "runs": [
      "text:135"
     ],
     "notes": "GLiNER2 server, self-hosted"
    },
    {
     "repo": "LiquidAI/LFM2.5-8B-A1B",
     "commit": "dd83f5c",
     "name": "LFM2.5-8B-A1B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-8b-a1b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 8.47,
     "params_active": 1.69,
     "quantization": "BF16",
     "score_0_100": -1.2483787289234785,
     "raw": {
      "metric": "accuracy",
      "value": 0.19001297016861218,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1288"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-8B-A1B@dd83f5c), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.05, seed 0, temperature 0.2, top_k 80; output cap 32,768"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya",
     "tag": "",
     "model_key": "laya",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.42,
     "params_active": 0.42,
     "quantization": null,
     "score_0_100": -2.221141374837875,
     "raw": {
      "metric": "accuracy",
      "value": 0.182230869001297,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.1491920233463036,
     "improvement_points": null,
     "runs": [
      "text:51"
     ],
     "notes": "Laya server, self-hosted"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya Multilingual",
     "tag": "",
     "model_key": "laya-multilingual",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.32,
     "params_active": 0.32,
     "quantization": null,
     "score_0_100": -2.221141374837875,
     "raw": {
      "metric": "accuracy",
      "value": 0.182230869001297,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.26038787289234744,
     "improvement_points": null,
     "runs": [
      "text:46"
     ],
     "notes": "Laya server, self-hosted"
    },
    {
     "repo": "fastino/gliner2.5-multi-v1",
     "commit": "2ca71aa",
     "name": "GLiNER2.5 Multi",
     "tag": "",
     "model_key": "gliner2.5-multi-base",
     "maker": "Fastino",
     "kind": "chat",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": -3.0317769130998733,
     "raw": {
      "metric": "accuracy",
      "value": 0.17574578469520102,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": 0.35828060109830245,
     "improvement_points": null,
     "runs": [
      "text:122"
     ],
     "notes": "GLiNER2 server, self-hosted"
    },
    {
     "repo": "openbmb/MiniCPM5-2B",
     "commit": "42433e9",
     "name": "MiniCPM5-2B",
     "tag": "reasoning off · chat",
     "model_key": "minicpm5-2b",
     "maker": "OpenBMB",
     "kind": "chat",
     "params_total": 2.52,
     "params_active": 2.52,
     "quantization": "BF16",
     "score_0_100": -18.51491569390402,
     "raw": {
      "metric": "accuracy",
      "value": 0.05188067444876784,
      "questions": 1542,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "text:1299"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision openbmb/MiniCPM5-2B@42433e9), one question per request, plain answer line CEVAP: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking off; output cap 32,768"
    }
   ]
  },
  {
   "bench": "reading-english",
   "title": "Reading · English",
   "url": "https://zekiworks.github.io/analysis/decision-models/english/",
   "updated": "2026-11-16T20:50:26Z",
   "score": {
    "zero": "chance: 1 of 5 options right on every question (accuracy 20%)",
    "hundred": "every one of the 547 kept reading questions, translated into English, right (accuracy 100%)"
   },
   "models": [
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.2",
     "tag": "max · chat",
     "model_key": "claude-fable-5.2",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 97.48628884826324,
     "raw": {
      "metric": "accuracy",
      "value": 0.979890310786106,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.10658135283363845,
     "improvement_points": null,
     "runs": [
      "english:971"
     ],
     "notes": "Claude Code CLI, Claude subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Hydrogen",
     "tag": "high · chat",
     "model_key": "gemini-4-hydrogen",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.88665447897623,
     "raw": {
      "metric": "accuracy",
      "value": 0.9670932358318098,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.030840950639851206,
     "improvement_points": null,
     "runs": [
      "english:1107"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Astra",
     "tag": "high · chat",
     "model_key": "gpt-6-astra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.20109689213892,
     "raw": {
      "metric": "accuracy",
      "value": 0.9616087751371115,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.018352833638030697,
     "improvement_points": null,
     "runs": [
      "english:615"
     ],
     "notes": "Codex CLI, ChatGPT subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Terra",
     "tag": "high · chat",
     "model_key": "gpt-6.1-terra",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 95.20109689213892,
     "raw": {
      "metric": "accuracy",
      "value": 0.9616087751371115,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.012541133455215183,
     "improvement_points": null,
     "runs": [
      "english:620"
     ],
     "notes": "Codex CLI, ChatGPT subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 3.8 Flash",
     "tag": "high · chat",
     "model_key": "gemini-3.8-flash",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 94.9725776965265,
     "raw": {
      "metric": "accuracy",
      "value": 0.9597806215722121,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.016526508226690197,
     "improvement_points": null,
     "runs": [
      "english:583"
     ],
     "notes": "Gemini API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Argon",
     "tag": "low · chat",
     "model_key": "gemini-4-argon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 94.9725776965265,
     "raw": {
      "metric": "accuracy",
      "value": 0.9597806215722121,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0152650822669123,
     "improvement_points": null,
     "runs": [
      "english:593"
     ],
     "notes": "Gemini API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Fable 5.1",
     "tag": "high · chat",
     "model_key": "claude-fable-5.1",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 94.9725776965265,
     "raw": {
      "metric": "accuracy",
      "value": 0.9597806215722121,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.11833638025594223,
     "improvement_points": null,
     "runs": [
      "english:580"
     ],
     "notes": "Claude Code CLI, Claude subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6.1 Sol",
     "tag": "high · chat",
     "model_key": "gpt-6.1-sol",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 94.51553930530166,
     "raw": {
      "metric": "accuracy",
      "value": 0.9561243144424132,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02412248628885336,
     "improvement_points": null,
     "runs": [
      "english:618"
     ],
     "notes": "Codex CLI, ChatGPT subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Opus 5.5",
     "tag": "high · chat",
     "model_key": "claude-opus-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 93.14442413162705,
     "raw": {
      "metric": "accuracy",
      "value": 0.9451553930530164,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.13327239488117043,
     "improvement_points": null,
     "runs": [
      "english:587"
     ],
     "notes": "Claude Code CLI, Claude subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen4-Max",
     "tag": "high · chat",
     "model_key": "qwen4-max",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 93.14442413162705,
     "raw": {
      "metric": "accuracy",
      "value": 0.9451553930530164,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08451553930530313,
     "improvement_points": null,
     "runs": [
      "english:1334"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint (the maker's), fallbacks off, reasoning effort high, output cap 32,768 tokens, reasoning included; the answer schema; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini 4 Carbon",
     "tag": "high · chat",
     "model_key": "gemini-4-carbon",
     "maker": "Google",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 92.91590493601461,
     "raw": {
      "metric": "accuracy",
      "value": 0.943327239488117,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03141242937852733,
     "improvement_points": null,
     "runs": [
      "english:1055"
     ],
     "notes": "Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen1.5",
     "commit": "8e56aa1b236a6dd1064dcfd22ba1e5e360e4c140",
     "name": "KnowLine-31B-Gen1.5",
     "tag": "",
     "model_key": "knowline-31b-gen1.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 92.00182815356489,
     "raw": {
      "metric": "accuracy",
      "value": 0.9360146252285192,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05000751445688127,
     "improvement_points": null,
     "runs": [
      "english:953"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Grok 4.7",
     "tag": "high · chat",
     "model_key": "grok-4.7",
     "maker": "xAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 91.77330895795247,
     "raw": {
      "metric": "accuracy",
      "value": 0.9341864716636198,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.11109689213893964,
     "improvement_points": null,
     "runs": [
      "english:579"
     ],
     "notes": "xAI Responses API, SuperGrok subscription; the kept reading questions in English"
    },
    {
     "repo": "Poindexter-Labs/Copernicus-31B",
     "commit": "be71525",
     "name": "Copernicus-31B",
     "tag": "",
     "model_key": "copernicus-31b",
     "maker": "Poindexter Labs",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 91.08775137111516,
     "raw": {
      "metric": "accuracy",
      "value": 0.9287020109689214,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0779339001233629,
     "improvement_points": null,
     "runs": [
      "english:1168"
     ],
     "notes": "Copernicus server (the card's copernicus_server.py, Transformers), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Haiku 5.5",
     "tag": "max · chat",
     "model_key": "claude-haiku-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 90.85923217550274,
     "raw": {
      "metric": "accuracy",
      "value": 0.926873857404022,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.18385740402193793,
     "improvement_points": null,
     "runs": [
      "english:949"
     ],
     "notes": "OpenRouter chat API, Anthropic's own endpoint, fallbacks off, output cap 128,000 tokens, reasoning max; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "DeepSeek-V4.1-Flash",
     "tag": "reasoning on · chat · 1017",
     "model_key": "deepseek-v4.1-flash-1017",
     "maker": "DeepSeek",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": "FP4 experts, FP8 and BF16 layers",
     "score_0_100": 90.6307129798903,
     "raw": {
      "metric": "accuracy",
      "value": 0.9250457038391224,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.07494203009143319,
     "improvement_points": null,
     "runs": [
      "english:808"
     ],
     "notes": "SGLang, self-hosted, published mixed precision; 1017 weights (revision 1bae0e1); the kept reading questions in English"
    },
    {
     "repo": "decisionmodels/REAX-1.3",
     "commit": "fa4bf7e",
     "name": "REAX 1.3",
     "tag": "",
     "model_key": "reax-1.3",
     "maker": "REAX",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 90.6307129798903,
     "raw": {
      "metric": "accuracy",
      "value": 0.9250457038391224,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03601321395054697,
     "improvement_points": null,
     "runs": [
      "english:829"
     ],
     "notes": "REAX Engine server (reax13 mode, the maker's release code), self-hosted, FP8; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Kimi K3.1 Pro",
     "tag": "low · chat",
     "model_key": "kimi-k3.1-pro",
     "maker": "Moonshot",
     "kind": "chat",
     "params_total": 2800.0,
     "params_active": 104.0,
     "quantization": null,
     "score_0_100": 90.40219378427788,
     "raw": {
      "metric": "accuracy",
      "value": 0.923217550274223,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05718464351006046,
     "improvement_points": null,
     "runs": [
      "english:584"
     ],
     "notes": "OpenRouter chat API, Moonshot AI's endpoint (mxfp4), fallbacks off; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen2",
     "commit": "fd99a9e",
     "name": "KnowLine-31B-Gen2",
     "tag": "",
     "model_key": "knowline-31b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 90.40219378427788,
     "raw": {
      "metric": "accuracy",
      "value": 0.923217550274223,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.07692675098310478,
     "improvement_points": null,
     "runs": [
      "english:1327"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "MachinePulseAI/karpo-decision-31B",
     "commit": "d69cdd5",
     "name": "Karpo Decision 31B",
     "tag": "",
     "model_key": "karpo-decision-31b",
     "maker": "MachinePulse",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 90.17367458866545,
     "raw": {
      "metric": "accuracy",
      "value": 0.9213893967093236,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0238317259997867,
     "improvement_points": null,
     "runs": [
      "english:1159"
     ],
     "notes": "karpo-decision server (the System One server of the card's karpo-decision package, machinepulse-ai/karpo-decision at ac40d39), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.5-397B-A17B",
     "tag": "high · chat",
     "model_key": "qwen-3.5-397b-a17b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 397.0,
     "params_active": 17.0,
     "quantization": null,
     "score_0_100": 89.25959780621572,
     "raw": {
      "metric": "accuracy",
      "value": 0.9140767824497258,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03308957952468328,
     "improvement_points": null,
     "runs": [
      "english:871"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini Decision 1",
     "tag": "20261111",
     "model_key": "gemini-decision-1",
     "maker": "Google",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 88.80255941499084,
     "raw": {
      "metric": "accuracy",
      "value": 0.9104204753199269,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04488473464044773,
     "improvement_points": null,
     "runs": [
      "english:1009"
     ],
     "notes": "OpenRouter decisions API, Google AI Studio endpoint; the kept reading questions in English"
    },
    {
     "repo": "logitshq/Logits-Sigma-1-Duo",
     "commit": "a0c1b6e",
     "name": "Logits Sigma-1 Duo",
     "tag": "",
     "model_key": "logits-sigma-1-duo",
     "maker": "Logits",
     "kind": "decision",
     "params_total": 57.36,
     "params_active": 57.36,
     "quantization": "FP8",
     "score_0_100": 88.80255941499084,
     "raw": {
      "metric": "accuracy",
      "value": 0.9104204753199269,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.07979727737554777,
     "improvement_points": null,
     "runs": [
      "english:1229"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiDE-27B",
     "commit": "4d32743",
     "name": "GLiDE 27B",
     "tag": "",
     "model_key": "glide-27b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 88.57404021937842,
     "raw": {
      "metric": "accuracy",
      "value": 0.9085923217550275,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.01697024716258676,
     "improvement_points": null,
     "runs": [
      "english:901"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "turion-ai/turion-jev-31b",
     "commit": "866c0fc6e040cba23fcc72113a80e5c99d5ebab1",
     "name": "Turion Jev 31B",
     "tag": "",
     "model_key": "turion-jev-31b",
     "maker": "Turion.AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 at load",
     "score_0_100": 88.345521023766,
     "raw": {
      "metric": "accuracy",
      "value": 0.906764168190128,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.11628131444241334,
     "improvement_points": null,
     "runs": [
      "english:1075"
     ],
     "notes": "Turion Jev server (turion_jev.serve on vLLM, the maker's release code), self-hosted, FP8 at load; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1-ultra",
     "tag": "",
     "model_key": "liquid-d1-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 88.345521023766,
     "raw": {
      "metric": "accuracy",
      "value": 0.906764168190128,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04387182824822371,
     "improvement_points": null,
     "runs": [
      "english:411",
      "english:418",
      "english:422",
      "english:426",
      "english:428"
     ],
     "notes": "Liquid AI decisions API; the kept reading questions in English"
    },
    {
     "repo": "sthanika-ai/Sieve-31B-v3",
     "commit": "8fcee858f7f20a50bfa3c8f984c22adb3cbb46a2",
     "name": "Sieve 31B v3",
     "tag": "",
     "model_key": "sieve-31b-v3",
     "maker": "Sthānika AI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 88.345521023766,
     "raw": {
      "metric": "accuracy",
      "value": 0.906764168190128,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04257458866544805,
     "improvement_points": null,
     "runs": [
      "english:1064"
     ],
     "notes": "Sieve server (sieve-serve, the maker's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "google/gemma-4-31B-it",
     "commit": "842da37",
     "name": "Gemma 4 31B",
     "tag": "reasoning off · chat",
     "model_key": "gemma-4-31b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "FP8 weights and KV cache",
     "score_0_100": 88.11700182815358,
     "raw": {
      "metric": "accuracy",
      "value": 0.9049360146252285,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08489003108485879,
     "improvement_points": null,
     "runs": [
      "english:774"
     ],
     "notes": "vLLM, self-hosted, FP8; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1.2",
     "tag": "API · with web",
     "model_key": "perplexity-decider-27b-v1.2-with-web",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 88.11700182815358,
     "raw": {
      "metric": "accuracy",
      "value": 0.9049360146252285,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0364576720167722,
     "improvement_points": null,
     "runs": [
      "english:639"
     ],
     "notes": "Perplexity Decisions API; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiDE",
     "commit": "a7e4589",
     "name": "GLiDE",
     "tag": "v1.2",
     "model_key": "fastino-glide-v1.2",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 87.88848263254113,
     "raw": {
      "metric": "accuracy",
      "value": 0.903107861060329,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02740066195052254,
     "improvement_points": null,
     "runs": [
      "english:521"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2-ultra",
     "tag": "",
     "model_key": "liquid-d1.2-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 87.88848263254113,
     "raw": {
      "metric": "accuracy",
      "value": 0.903107861060329,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08484512831598903,
     "improvement_points": null,
     "runs": [
      "english:1359"
     ],
     "notes": "Liquid AI decisions API; the kept reading questions in English"
    },
    {
     "repo": "baselabs/mammoth-1",
     "commit": "61cec44",
     "name": "Mammoth-1",
     "tag": "",
     "model_key": "mammoth-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 87.65996343692869,
     "raw": {
      "metric": "accuracy",
      "value": 0.9012797074954296,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.024374643510054157,
     "improvement_points": null,
     "runs": [
      "english:503"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "vllm-sr/d3",
     "commit": "20f37b9",
     "name": "d3",
     "tag": "",
     "model_key": "vllm-sr-d3",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 87.20292504570384,
     "raw": {
      "metric": "accuracy",
      "value": 0.8976234003656307,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03550667558306448,
     "improvement_points": null,
     "runs": [
      "english:554"
     ],
     "notes": "vLLM System One endpoint (nightly), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Muse Ember",
     "tag": "max · chat",
     "model_key": "muse-ember",
     "maker": "Meta",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 86.97440585009142,
     "raw": {
      "metric": "accuracy",
      "value": 0.8957952468007313,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.07005484460694747,
     "improvement_points": null,
     "runs": [
      "english:942"
     ],
     "notes": "OpenRouter chat API, Meta's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning max; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3-Flash",
     "tag": "reasoning on · chat",
     "model_key": "glm-5.3-flash",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 320.0,
     "params_active": 18.0,
     "quantization": "EXL3 4-bit experts, FP8 dense layers and KV cache",
     "score_0_100": 86.97440585009142,
     "raw": {
      "metric": "accuracy",
      "value": 0.8957952468007313,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.06965265082267022,
     "improvement_points": null,
     "runs": [
      "english:823"
     ],
     "notes": "TensorFold, self-hosted, EXL3 4-bit experts; the kept reading questions in English"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.3",
     "commit": "9c2ebb2",
     "name": "Perplexity Decider 27B v1.3",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 86.97440585009142,
     "raw": {
      "metric": "accuracy",
      "value": 0.8957952468007313,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.033677202614917454,
     "improvement_points": null,
     "runs": [
      "english:1137"
     ],
     "notes": "Perplexity Decider server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "nace-ai/drex-max-v1",
     "commit": "cac65e1",
     "name": "Drex Max",
     "tag": "",
     "model_key": "drex-max-v1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 86.97440585009142,
     "raw": {
      "metric": "accuracy",
      "value": 0.8957952468007313,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.026342953016338093,
     "improvement_points": null,
     "runs": [
      "english:552"
     ],
     "notes": "Drex Max server (Nace's loader, drex_max), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "vllm-sr/d3-ultra",
     "commit": "52ba9e7",
     "name": "d3-ultra",
     "tag": "",
     "model_key": "vllm-sr-d3-ultra",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 86.74588665447897,
     "raw": {
      "metric": "accuracy",
      "value": 0.8939670932358318,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02817219202914336,
     "improvement_points": null,
     "runs": [
      "english:550"
     ],
     "notes": "d3 package server (transformers, the release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "Qwen/Qwen3.8-27B",
     "commit": "1d4bf0f",
     "name": "Qwen3.8-27B",
     "tag": "low · chat",
     "model_key": "qwen3.8-27b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 27.78,
     "params_active": 27.78,
     "quantization": "BF16",
     "score_0_100": 86.28884826325411,
     "raw": {
      "metric": "accuracy",
      "value": 0.8903107861060329,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.10960974413925095,
     "improvement_points": null,
     "runs": [
      "english:667"
     ],
     "notes": "vLLM, self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiMo-V2.6-Flash",
     "tag": "high · chat",
     "model_key": "mimo-v2.6-flash",
     "maker": "Xiaomi",
     "kind": "chat",
     "params_total": 309.0,
     "params_active": 15.0,
     "quantization": null,
     "score_0_100": 86.28884826325411,
     "raw": {
      "metric": "accuracy",
      "value": 0.8903107861060329,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0647897623400371,
     "improvement_points": null,
     "runs": [
      "english:878"
     ],
     "notes": "OpenRouter chat API, Xiaomi's own endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2",
     "tag": "",
     "model_key": "liquid-d1.2",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 86.28884826325411,
     "raw": {
      "metric": "accuracy",
      "value": 0.8903107861060329,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08276620031021971,
     "improvement_points": null,
     "runs": [
      "english:1357"
     ],
     "notes": "Liquid AI decisions API; the kept reading questions in English"
    },
    {
     "repo": "perplexity-ai/pplx-decider-v1.2",
     "commit": "225c913",
     "name": "Perplexity Decider 27B v1.2",
     "tag": "DGX Spark",
     "model_key": "perplexity-decider-27b-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 85.83180987202923,
     "raw": {
      "metric": "accuracy",
      "value": 0.886654478976234,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03733364820187485,
     "improvement_points": null,
     "runs": [
      "english:527"
     ],
     "notes": "Perplexity Decider server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex Max v1.1",
     "tag": "",
     "model_key": "drex-max-1.1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 85.6032906764168,
     "raw": {
      "metric": "accuracy",
      "value": 0.8848263254113345,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03773080842131497,
     "improvement_points": null,
     "runs": [
      "english:399"
     ],
     "notes": "Nace.AI System One API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Perplexity Decider 27B v1",
     "tag": "",
     "model_key": "perplexity-decider-27b-v1",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": null,
     "score_0_100": 85.6032906764168,
     "raw": {
      "metric": "accuracy",
      "value": 0.8848263254113345,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02619702423757918,
     "improvement_points": null,
     "runs": [
      "english:626",
      "english:627",
      "english:628",
      "english:629",
      "english:630"
     ],
     "notes": "Perplexity Decisions API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Inkling-Small",
     "tag": "high · chat · Jul 2026",
     "model_key": "inkling-small",
     "maker": "Thinking Machines",
     "kind": "chat",
     "params_total": 265.96,
     "params_active": 12.0,
     "quantization": null,
     "score_0_100": 85.37477148080438,
     "raw": {
      "metric": "accuracy",
      "value": 0.8829981718464351,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02574040219378395,
     "improvement_points": null,
     "runs": [
      "english:867"
     ],
     "notes": "OpenRouter chat API, DeepInfra's endpoint (fp8), fallbacks off, output cap 65,536 tokens, reasoning included; no response format (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.1",
     "tag": "",
     "model_key": "liquid-d1.1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 85.37477148080438,
     "raw": {
      "metric": "accuracy",
      "value": 0.8829981718464351,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04150821475677424,
     "improvement_points": null,
     "runs": [
      "english:565",
      "english:568",
      "english:570",
      "english:573",
      "english:575"
     ],
     "notes": "Liquid AI decisions API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GLM-5.3",
     "tag": "low · chat",
     "model_key": "glm-5.3",
     "maker": "Z.ai",
     "kind": "chat",
     "params_total": 753.33,
     "params_active": 753.33,
     "quantization": null,
     "score_0_100": 84.68921389396708,
     "raw": {
      "metric": "accuracy",
      "value": 0.8775137111517367,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03590493601462523,
     "improvement_points": null,
     "runs": [
      "english:577"
     ],
     "notes": "OpenRouter chat API, Fireworks endpoint, fallbacks off; the kept reading questions in English"
    },
    {
     "repo": "tattrongvu/Jeffan-9B-1211",
     "commit": "862d1a7",
     "name": "Jeffan-9B-1211",
     "tag": "",
     "model_key": "jeffan-9b-1211",
     "maker": "tattrongvu",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 84.46069469835466,
     "raw": {
      "metric": "accuracy",
      "value": 0.8756855575868373,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08882281439908829,
     "improvement_points": null,
     "runs": [
      "english:1195"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Qwen3.8 Flash",
     "tag": "no reasoning · chat",
     "model_key": "qwen3.8-flash",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 83.77513711151737,
     "raw": {
      "metric": "accuracy",
      "value": 0.870201096892139,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05471663619743793,
     "improvement_points": null,
     "runs": [
      "english:986"
     ],
     "notes": "OpenRouter chat API, Alibaba's endpoint, fallbacks off, reasoning off; the kept reading questions in English"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning on · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 83.77513711151737,
     "raw": {
      "metric": "accuracy",
      "value": 0.870201096892139,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1072"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking on; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1",
     "tag": "",
     "model_key": "liquid-d1",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 83.54661791590492,
     "raw": {
      "metric": "accuracy",
      "value": 0.8683729433272395,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02560456628254348,
     "improvement_points": null,
     "runs": [
      "english:430",
      "english:432",
      "english:433",
      "english:434",
      "english:435"
     ],
     "notes": "Liquid AI decisions API; the kept reading questions in English"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-9B-Instant",
     "commit": "18a17f734966b213b2f00fa9c915e8b9af2a9b20",
     "name": "StartLux-Decision-v1.1-9B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-9b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 83.54661791590492,
     "raw": {
      "metric": "accuracy",
      "value": 0.8683729433272395,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0658049895185053,
     "improvement_points": null,
     "runs": [
      "english:1293"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "google/gemma-4-12B-it",
     "commit": "43c9d01",
     "name": "Gemma 4 12B Unified",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-12b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 83.54661791590492,
     "raw": {
      "metric": "accuracy",
      "value": 0.8683729433272395,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:895"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-12B-it@43c9d01), one question per request, plain answer line ANSWER: X (no answer schema); sampling temperature 0.0; output cap 8,192; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-12B-Gen1",
     "commit": "f26fbc2",
     "name": "KnowLine-12B-Gen1",
     "tag": "",
     "model_key": "knowline-12b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 83.3180987202925,
     "raw": {
      "metric": "accuracy",
      "value": 0.8665447897623401,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.11124588259520785,
     "improvement_points": null,
     "runs": [
      "english:1313"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "Cloudflare/clef",
     "commit": "2f3de3d",
     "name": "Clef",
     "tag": "",
     "model_key": "clef",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 27.48,
     "params_active": 27.48,
     "quantization": "BF16",
     "score_0_100": 83.3180987202925,
     "raw": {
      "metric": "accuracy",
      "value": 0.8665447897623401,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03148500914076833,
     "improvement_points": null,
     "runs": [
      "english:613"
     ],
     "notes": "Clef server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "low · chat",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.86106032906766,
     "raw": {
      "metric": "accuracy",
      "value": 0.8628884826325411,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.10084095063985805,
     "improvement_points": null,
     "runs": [
      "english:617"
     ],
     "notes": "Codex CLI, ChatGPT subscription; the kept reading questions in English"
    },
    {
     "repo": "OmniJev/OneJev-27B",
     "commit": "a5e5f55",
     "name": "OneJev-27B",
     "tag": "",
     "model_key": "onejev-27b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 82.86106032906766,
     "raw": {
      "metric": "accuracy",
      "value": 0.8628884826325411,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0972213546617914,
     "improvement_points": null,
     "runs": [
      "english:778"
     ],
     "notes": "OneJev server (qev, OmniJev's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiDE-9B",
     "commit": "ab54df5",
     "name": "GLiDE 9B",
     "tag": "",
     "model_key": "glide-9b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 7.94,
     "params_active": 7.94,
     "quantization": "BF16",
     "score_0_100": 82.86106032906766,
     "raw": {
      "metric": "accuracy",
      "value": 0.8628884826325411,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.055857307953141797,
     "improvement_points": null,
     "runs": [
      "english:909"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Claude Sonnet 5.5",
     "tag": "high · chat",
     "model_key": "claude-sonnet-5.5",
     "maker": "Anthropic",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.63254113345519,
     "raw": {
      "metric": "accuracy",
      "value": 0.8610603290676416,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.13868372943327234,
     "improvement_points": null,
     "runs": [
      "english:590"
     ],
     "notes": "Claude Code CLI, Claude subscription; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.14",
     "tag": "",
     "model_key": "jev-1.14.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.63254113345519,
     "raw": {
      "metric": "accuracy",
      "value": 0.8610603290676416,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02215722120658174,
     "improvement_points": null,
     "runs": [
      "english:410",
      "english:416",
      "english:421",
      "english:423",
      "english:425"
     ],
     "notes": "TypeSafe System One API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap Preview",
     "tag": "",
     "model_key": "subq-snap-preview",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.40402193784277,
     "raw": {
      "metric": "accuracy",
      "value": 0.8592321755027422,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03225577310735607,
     "improvement_points": null,
     "runs": [
      "english:595",
      "english:597",
      "english:598",
      "english:599",
      "english:600"
     ],
     "notes": "SubQ Decisions API (public preview); the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap 1.1",
     "tag": "",
     "model_key": "subq-snap-1.1",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.40402193784277,
     "raw": {
      "metric": "accuracy",
      "value": 0.8592321755027422,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03225577310735607,
     "improvement_points": null,
     "runs": [
      "english:413"
     ],
     "notes": "SubQ Decisions API (public preview); the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.13",
     "tag": "",
     "model_key": "jev-1.13.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.40402193784277,
     "raw": {
      "metric": "accuracy",
      "value": 0.8592321755027422,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.030859232175503442,
     "improvement_points": null,
     "runs": [
      "english:564",
      "english:567",
      "english:569",
      "english:572",
      "english:574"
     ],
     "notes": "TypeSafe System One API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex 2.0",
     "tag": "",
     "model_key": "drex-2.0",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 35.11,
     "params_active": 3.0,
     "quantization": null,
     "score_0_100": 82.40402193784277,
     "raw": {
      "metric": "accuracy",
      "value": 0.8592321755027422,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.06914899451553924,
     "improvement_points": null,
     "runs": [
      "english:566",
      "english:571",
      "english:576",
      "english:581",
      "english:586"
     ],
     "notes": "Nace.AI System One API; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen1",
     "commit": "325378d",
     "name": "KnowLine-9B-Gen1",
     "tag": "",
     "model_key": "knowline-9b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 82.40402193784277,
     "raw": {
      "metric": "accuracy",
      "value": 0.8592321755027422,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0346777326975474,
     "improvement_points": null,
     "runs": [
      "english:1177"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "apertixai/cerno-9b-v1.2",
     "commit": "f02a632fbe3a4698cd1ff5c0b929ea240e2b8eda",
     "name": "Cerno 9B v1.2",
     "tag": "",
     "model_key": "cerno-9b-v1.2",
     "maker": "Apertix AI",
     "kind": "decision",
     "params_total": 8.95,
     "params_active": 8.95,
     "quantization": "BF16",
     "score_0_100": 82.40402193784277,
     "raw": {
      "metric": "accuracy",
      "value": 0.8592321755027422,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04191992687385757,
     "improvement_points": null,
     "runs": [
      "english:1085"
     ],
     "notes": "Cerno server (a thin System One server around the card's CernoEngine, transformers: logic's for its English runs, compactor's di_serve.py for its Turkish runs), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Mercury Decide",
     "tag": "",
     "model_key": "inception-mercury-decide",
     "maker": "Inception",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.17550274223035,
     "raw": {
      "metric": "accuracy",
      "value": 0.8574040219378428,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.06641069949989642,
     "improvement_points": null,
     "runs": [
      "english:398",
      "english:400",
      "english:401",
      "english:402",
      "english:403"
     ],
     "notes": "OpenRouter System One API; the kept reading questions in English"
    },
    {
     "repo": "newsbang/adjudex-9b-v5",
     "commit": "ffdcb82",
     "name": "Adjudex 9B v5",
     "tag": "",
     "model_key": "adjudex-9b-v5",
     "maker": "newsbang",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 82.17550274223035,
     "raw": {
      "metric": "accuracy",
      "value": 0.8574040219378428,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03432343925289232,
     "improvement_points": null,
     "runs": [
      "english:1186"
     ],
     "notes": "Perplexity Decider server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Microsoft-Decision-1",
     "tag": "",
     "model_key": "microsoft-decision-1",
     "maker": "Microsoft",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.17550274223035,
     "raw": {
      "metric": "accuracy",
      "value": 0.8574040219378428,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.027993635427653724,
     "improvement_points": null,
     "runs": [
      "english:412",
      "english:420",
      "english:424",
      "english:427",
      "english:431"
     ],
     "notes": "OpenRouter System One API; the kept reading questions in English"
    },
    {
     "repo": "ZefanCai/Open-Jev-27B-v1.1",
     "commit": "28cf730",
     "name": "Open-Jev-27B-v1.1",
     "tag": "",
     "model_key": "open-jev-27b-v1.1",
     "maker": "Zefan Cai",
     "kind": "decision",
     "params_total": 27.8,
     "params_active": 27.8,
     "quantization": "BF16",
     "score_0_100": 81.7184643510055,
     "raw": {
      "metric": "accuracy",
      "value": 0.8537477148080439,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.45275189164504165,
     "improvement_points": null,
     "runs": [
      "english:760"
     ],
     "notes": "Open-Jev server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "baselabs/elephant-1",
     "commit": "b3f8a93",
     "name": "Elephant-1",
     "tag": "",
     "model_key": "elephant-1",
     "maker": "Base Labs",
     "kind": "decision",
     "params_total": 9.41,
     "params_active": 9.41,
     "quantization": "BF16",
     "score_0_100": 81.7184643510055,
     "raw": {
      "metric": "accuracy",
      "value": 0.8537477148080439,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.02687023217550313,
     "improvement_points": null,
     "runs": [
      "english:499"
     ],
     "notes": "Base Labs server (qev, the model card's loader), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-9B-Gen2",
     "commit": "8ec8eb6",
     "name": "KnowLine-9B-Gen2",
     "tag": "",
     "model_key": "knowline-9b-gen2",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 9.65,
     "params_active": 9.65,
     "quantization": "BF16",
     "score_0_100": 81.48994515539304,
     "raw": {
      "metric": "accuracy",
      "value": 0.8519195612431444,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08699633358497873,
     "improvement_points": null,
     "runs": [
      "english:1321"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang, PelaAI's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "MiniMax-M3.1-Flash",
     "tag": "high · chat",
     "model_key": "minimax-m3.1-flash",
     "maker": "MiniMax",
     "kind": "chat",
     "params_total": 428.0,
     "params_active": 23.0,
     "quantization": null,
     "score_0_100": 80.57586837294333,
     "raw": {
      "metric": "accuracy",
      "value": 0.8446069469835467,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.09040293040293255,
     "improvement_points": null,
     "runs": [
      "english:863"
     ],
     "notes": "OpenRouter chat API, MiniMax's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning included; JSON mode without the answer schema (the endpoint has no structured outputs), so the prompt adds a paragraph stating the answer layout; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "Decisions API",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 80.11882998171845,
     "raw": {
      "metric": "accuracy",
      "value": 0.8409506398537477,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03190825688073419,
     "improvement_points": null,
     "runs": [
      "english:397"
     ],
     "notes": "OpenAI Decisions API (public beta); the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "celeris-1-decision",
     "tag": "",
     "model_key": "celeris-1-decision",
     "maker": "Celeris",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 80.11882998171845,
     "raw": {
      "metric": "accuracy",
      "value": 0.8409506398537477,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05191956124314509,
     "improvement_points": null,
     "runs": [
      "english:546"
     ],
     "notes": "Celeris System One API; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Cygnet",
     "tag": "",
     "model_key": "cygnet",
     "maker": "Blockbrain Labs",
     "kind": "decision",
     "params_total": 11.96,
     "params_active": 11.96,
     "quantization": "BF16",
     "score_0_100": 79.89031078610603,
     "raw": {
      "metric": "accuracy",
      "value": 0.8391224862888482,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04304977944568373,
     "improvement_points": null,
     "runs": [
      "english:753"
     ],
     "notes": "Cygnet decision server on vLLM, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "EldanRing/Winnow-12B",
     "commit": "b2b1421",
     "name": "Winnow-12B",
     "tag": "",
     "model_key": "winnow-12b",
     "maker": "EldanRing",
     "kind": "decision",
     "params_total": 11.91,
     "params_active": 11.91,
     "quantization": "Q8_0",
     "score_0_100": 79.89031078610603,
     "raw": {
      "metric": "accuracy",
      "value": 0.8391224862888482,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08650655823227467,
     "improvement_points": null,
     "runs": [
      "english:746"
     ],
     "notes": "Winnow server (llama.cpp), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.3",
     "commit": "d06fb3f",
     "name": "Perplexity Decider Flash v1.3",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.3",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 79.6617915904936,
     "raw": {
      "metric": "accuracy",
      "value": 0.8372943327239488,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.026805325207281724,
     "improvement_points": null,
     "runs": [
      "english:1126"
     ],
     "notes": "Perplexity Decider server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "Valen-Team/Valen-Decision-1.5-9B",
     "commit": "5a51211",
     "name": "Valen-Decision-1.5-9B",
     "tag": "",
     "model_key": "valen-decision-1.5-9b",
     "maker": "Valen-Team",
     "kind": "decision",
     "params_total": 8.4,
     "params_active": 8.4,
     "quantization": "BF16",
     "score_0_100": 79.6617915904936,
     "raw": {
      "metric": "accuracy",
      "value": 0.8372943327239488,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08429313076698099,
     "improvement_points": null,
     "runs": [
      "english:1278"
     ],
     "notes": "Valen server (the card's valen-vllm plugin, Valen-Team/Valen at 773d3a6, in vllm/vllm-openai v0.33.0), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "celeris-1.1-decision",
     "tag": "",
     "model_key": "celeris-1.1-decision",
     "maker": "Celeris",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 79.6617915904936,
     "raw": {
      "metric": "accuracy",
      "value": 0.8372943327239488,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.049597806215722774,
     "improvement_points": null,
     "runs": [
      "english:882"
     ],
     "notes": "Celeris System One API; the kept reading questions in English"
    },
    {
     "repo": "Patrick-18/Dexy-Decision-9B-V2",
     "commit": "4f434e2",
     "name": "Dexy-Decision-9B-V2",
     "tag": "",
     "model_key": "dexy-decision-9b-v2",
     "maker": "Patrick-18",
     "kind": "decision",
     "params_total": 8.39,
     "params_active": 8.39,
     "quantization": "BF16",
     "score_0_100": 79.43327239488114,
     "raw": {
      "metric": "accuracy",
      "value": 0.8354661791590493,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04177158883254661,
     "improvement_points": null,
     "runs": [
      "english:1258"
     ],
     "notes": "Dexy server (the card's Transformers d3_server.py, in its requirements.txt pins: torch 2.14.0, transformers 5.19.0), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "scar-ai/vera-medium-omni",
     "commit": "1f23a6c",
     "name": "vera-medium-omni",
     "tag": "",
     "model_key": "vera-medium-omni",
     "maker": "scar-ai",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 79.43327239488114,
     "raw": {
      "metric": "accuracy",
      "value": 0.8354661791590493,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.09859981718464357,
     "improvement_points": null,
     "runs": [
      "english:1267"
     ],
     "notes": "vera server (the card's own vera-serve from vera-s1 0.1.8, PyTorch engine), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-4B-Instant",
     "commit": "416088f77cfce09633906a3675c7cd2e5dabcf10",
     "name": "StartLux-Decision-v1.1-4B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-4b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 78.51919561243145,
     "raw": {
      "metric": "accuracy",
      "value": 0.8281535648994516,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03216870744173079,
     "improvement_points": null,
     "runs": [
      "english:1081"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "LiquidAI/LFM2.5-24B-A2B",
     "commit": "1769516",
     "name": "LFM2.5-24B-A2B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-24b-a2b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 23.97,
     "params_active": 2.45,
     "quantization": "BF16",
     "score_0_100": 78.290676416819,
     "raw": {
      "metric": "accuracy",
      "value": 0.8263254113345521,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1069"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-24B-A2B@1769516), one question per request, plain answer line ANSWER: X (no answer schema); sampling repetition_penalty 1.05, seed 0, temperature 0.1, top_k 50; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "google/gemma-4-E4B-it",
     "commit": "4ecc2ba",
     "name": "Gemma 4 E4B",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-e4b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 8.0,
     "params_active": 8.0,
     "quantization": "BF16",
     "score_0_100": 78.06215722120658,
     "raw": {
      "metric": "accuracy",
      "value": 0.8244972577696527,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:889"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E4B-it@4ecc2ba), one question per request, plain answer line CEVAP: X (no answer schema); sampling temperature 0.0; output cap 8,192; the kept reading questions in English"
    },
    {
     "repo": "victorbennett/Verdin-4B-v1.1",
     "commit": "45830d2",
     "name": "Verdin-4B-v1.1",
     "tag": "",
     "model_key": "verdin-4b-v1.1",
     "maker": "victorbennett",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 77.83363802559414,
     "raw": {
      "metric": "accuracy",
      "value": 0.8226691042047533,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03527083042969597,
     "improvement_points": null,
     "runs": [
      "english:1247"
     ],
     "notes": "Verdin server (the card's own python -m verdin.server, in the package versions its card lists), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "Cloudflare/clef-flash",
     "commit": "fde727a",
     "name": "Clef-Flash",
     "tag": "",
     "model_key": "clef-flash",
     "maker": "Cloudflare",
     "kind": "decision",
     "params_total": 9.53,
     "params_active": 9.53,
     "quantization": "BF16",
     "score_0_100": 77.83363802559414,
     "raw": {
      "metric": "accuracy",
      "value": 0.8226691042047533,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03996837294332693,
     "improvement_points": null,
     "runs": [
      "english:614"
     ],
     "notes": "Clef server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "perplexity-ai/pplx-decider-flash-v1.2",
     "commit": "6d771ad",
     "name": "Perplexity Decider Flash v1.2",
     "tag": "",
     "model_key": "perplexity-decider-flash-v1.2",
     "maker": "Perplexity",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 77.37659963436928,
     "raw": {
      "metric": "accuracy",
      "value": 0.8190127970749543,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.04133228190039147,
     "improvement_points": null,
     "runs": [
      "english:533"
     ],
     "notes": "Perplexity Decider server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-4B-Gen5.5",
     "commit": "779944d2cc8ffa19cef386aee83c83bcb967ade2",
     "name": "KnowLine 4B Gen5.5",
     "tag": "",
     "model_key": "knowline-4b-gen5.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "FP8 at load",
     "score_0_100": 76.00548446069469,
     "raw": {
      "metric": "accuracy",
      "value": 0.8080438756855576,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05798602865860706,
     "improvement_points": null,
     "runs": [
      "english:961"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, FP8 at load; the kept reading questions in English"
    },
    {
     "repo": "quincyqiang/chinese-decision-pointer-4b",
     "commit": "889f23e",
     "name": "Chinese Decision Pointer 4B",
     "tag": "",
     "model_key": "chinese-decision-pointer-4b",
     "maker": "quincyqiang",
     "kind": "decision",
     "params_total": 4.21,
     "params_active": 4.21,
     "quantization": "BF16",
     "score_0_100": 75.54844606946983,
     "raw": {
      "metric": "accuracy",
      "value": 0.8043875685557587,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03543935607162971,
     "improvement_points": null,
     "runs": [
      "english:1209"
     ],
     "notes": "Decision Index engine (compactor's di_serve.py around the engine its card ships), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "Qwen/Qwen3.5-4B",
     "commit": "851bf6e",
     "name": "Qwen3.5-4B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-4b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "BF16",
     "score_0_100": 75.54844606946983,
     "raw": {
      "metric": "accuracy",
      "value": 0.8043875685557587,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1143"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-4B@851bf6e), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "Qwen/Qwen3-14B",
     "commit": "40c0698",
     "name": "Qwen3-14B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3-14b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 74.6343692870201,
     "raw": {
      "metric": "accuracy",
      "value": 0.7970749542961609,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.17841177276527245,
     "improvement_points": null,
     "runs": [
      "english:770"
     ],
     "notes": "vLLM, self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "openbmb/MiniCPM5-2B",
     "commit": "42433e9",
     "name": "MiniCPM5-2B",
     "tag": "reasoning on · chat",
     "model_key": "minicpm5-2b",
     "maker": "OpenBMB",
     "kind": "chat",
     "params_total": 2.52,
     "params_active": 2.52,
     "quantization": "BF16",
     "score_0_100": 72.65386959171235,
     "raw": {
      "metric": "accuracy",
      "value": 0.7812309567336989,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1051",
      "english:1118",
      "english:1119"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision openbmb/MiniCPM5-2B@42433e9), one question per request, plain answer line ANSWER: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking on; output cap 32,768; the kept reading questions in English; the mean of 3 seed runs"
    },
    {
     "repo": "wayfind/metask-jev-4b-policy-mix",
     "commit": "ea20fe8",
     "name": "Metask-Jev-4B",
     "tag": "",
     "model_key": "metask-jev-4b",
     "maker": "Metask Lab",
     "kind": "decision",
     "params_total": 4.54,
     "params_active": 4.54,
     "quantization": "BF16",
     "score_0_100": 71.43510054844606,
     "raw": {
      "metric": "accuracy",
      "value": 0.7714808043875686,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.044228714524718896,
     "improvement_points": null,
     "runs": [
      "english:739"
     ],
     "notes": "Metask-Jev server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "google/gemma-4-E2B-it",
     "commit": "4adf157",
     "name": "Gemma 4 E2B",
     "tag": "reasoning on · chat",
     "model_key": "gemma-4-e2b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 5.12,
     "params_active": 5.12,
     "quantization": "BF16",
     "score_0_100": 70.29250457038391,
     "raw": {
      "metric": "accuracy",
      "value": 0.7623400365630713,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1349"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E2B-it@4adf157), one question per request, plain answer line ANSWER: X (no answer schema); sampling temperature 0.0, thinking on; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "Qwen/Qwen3.5-2B",
     "commit": "15852e8",
     "name": "Qwen3.5-2B",
     "tag": "reasoning on · chat",
     "model_key": "qwen3.5-2b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 69.83546617915906,
     "raw": {
      "metric": "accuracy",
      "value": 0.7586837294332724,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1341"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-2B@15852e8), one question per request, plain answer line ANSWER: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking on; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "google/gemma-4-E2B-it",
     "commit": "4adf157",
     "name": "Gemma 4 E2B",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-e2b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 5.12,
     "params_active": 5.12,
     "quantization": "BF16",
     "score_0_100": 68.23583180987202,
     "raw": {
      "metric": "accuracy",
      "value": 0.7458866544789763,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:892"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E2B-it@4adf157), one question per request, plain answer line ANSWER: X (no answer schema); sampling temperature 0.0; output cap 8,192; the kept reading questions in English"
    },
    {
     "repo": "ibm-granite/granite-4.2-3b",
     "commit": "4e9c275",
     "name": "Granite-4.2-3B",
     "tag": "reasoning on · chat",
     "model_key": "granite-4.2-3b",
     "maker": "IBM",
     "kind": "chat",
     "params_total": 3.66,
     "params_active": 3.66,
     "quantization": "BF16",
     "score_0_100": 68.00731261425959,
     "raw": {
      "metric": "accuracy",
      "value": 0.7440585009140768,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1104"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision ibm-granite/granite-4.2-3b@4e9c275), one question per request, plain answer line ANSWER: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking on; output cap 65,536; the kept reading questions in English"
    },
    {
     "repo": "LiquidAI/LFM2.5-2.6B",
     "commit": "7141b56",
     "name": "LFM2.5-2.6B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-2.6b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 2.7,
     "params_active": 2.7,
     "quantization": "BF16",
     "score_0_100": 67.3217550274223,
     "raw": {
      "metric": "accuracy",
      "value": 0.7385740402193784,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1034"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-2.6B@7141b56), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "ecloudtech/Erk-14B",
     "commit": "8584b14",
     "name": "Erk-14B",
     "tag": "reasoning off · chat",
     "model_key": "erk-14b",
     "maker": "eCloud Technology",
     "kind": "chat",
     "params_total": 14.77,
     "params_active": 14.77,
     "quantization": "BF16",
     "score_0_100": 65.03656307129798,
     "raw": {
      "metric": "accuracy",
      "value": 0.720292504570384,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.10872217547221544,
     "improvement_points": null,
     "runs": [
      "english:762"
     ],
     "notes": "vLLM, self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning off · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 64.7318708104814,
     "raw": {
      "metric": "accuracy",
      "value": 0.7178549664838513,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1223",
      "english:1302",
      "english:1303"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking off; output cap 32,768; the kept reading questions in English; the mean of 3 seed runs"
    },
    {
     "repo": "PelaAI/KnowLine-2B-Gen1",
     "commit": "53245d470a4ce316e32fd0d18c24e142c697f40b",
     "name": "KnowLine-2B-Gen1",
     "tag": "",
     "model_key": "knowline-2b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 64.12248628884825,
     "raw": {
      "metric": "accuracy",
      "value": 0.7129798903107861,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05134902695101977,
     "improvement_points": null,
     "runs": [
      "english:1057"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "LiquidAI/LFM3-1.7B",
     "commit": "0dc4268",
     "name": "LFM3-1.7B",
     "tag": "reasoning on · chat",
     "model_key": "lfm3-1.7b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 1.73,
     "params_active": 1.73,
     "quantization": "BF16",
     "score_0_100": 63.89396709323583,
     "raw": {
      "metric": "accuracy",
      "value": 0.7111517367458867,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1091",
      "english:1115",
      "english:1116"
     ],
     "notes": "our GPU (logic's server, vLLM nightly 0.34.0rc1.dev302 (commit be47189f; no released vLLM serves LFM3) BF16, revision LiquidAI/LFM3-1.7B@0dc4268), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, repetition_penalty 1.1, seed 0, temperature 0.1, top_k 50, top_p 1.0, thinking on; output cap 32,768; the kept reading questions in English; the mean of 3 seed runs"
    },
    {
     "repo": "LiquidAI/LFM2.5-8B-A1B",
     "commit": "dd83f5c",
     "name": "LFM2.5-8B-A1B",
     "tag": "default reasoning · chat",
     "model_key": "lfm2.5-8b-a1b",
     "maker": "Liquid AI",
     "kind": "chat",
     "params_total": 8.47,
     "params_active": 1.69,
     "quantization": "BF16",
     "score_0_100": 62.979890310786104,
     "raw": {
      "metric": "accuracy",
      "value": 0.7038391224862889,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:886"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision LiquidAI/LFM2.5-8B-A1B@dd83f5c), one question per request, plain answer line CEVAP: X (no answer schema); sampling repetition_penalty 1.05, seed 0, temperature 0.2, top_k 80; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiDE-2B",
     "commit": "00f122c",
     "name": "GLiDE 2B",
     "tag": "",
     "model_key": "glide-2b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 1.88,
     "params_active": 1.88,
     "quantization": "BF16",
     "score_0_100": 62.52285191956125,
     "raw": {
      "metric": "accuracy",
      "value": 0.70018281535649,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08180401882017321,
     "improvement_points": null,
     "runs": [
      "english:917"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-2B-Instant",
     "commit": "f1e6cd147dda82c9ffa3e8e725d5c1f175566fb1",
     "name": "StartLux-Decision-v1.1-2B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-2b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 62.06581352833638,
     "raw": {
      "metric": "accuracy",
      "value": 0.696526508226691,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08179573949711465,
     "improvement_points": null,
     "runs": [
      "english:973"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "openbmb/MiniCPM5-2B",
     "commit": "42433e9",
     "name": "MiniCPM5-2B",
     "tag": "reasoning off · chat",
     "model_key": "minicpm5-2b",
     "maker": "OpenBMB",
     "kind": "chat",
     "params_total": 2.52,
     "params_active": 2.52,
     "quantization": "BF16",
     "score_0_100": 59.704448507007925,
     "raw": {
      "metric": "accuracy",
      "value": 0.6776355880560634,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1094",
      "english:1300",
      "english:1301"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision openbmb/MiniCPM5-2B@42433e9), one question per request, plain answer line CEVAP: X (no answer schema); sampling seed 0, temperature 1.0, top_p 0.95, thinking off; output cap 32,768; the kept reading questions in English; the mean of 3 seed runs"
    },
    {
     "repo": "Qwen/Qwen3.5-2B",
     "commit": "15852e8",
     "name": "Qwen3.5-2B",
     "tag": "reasoning off · chat",
     "model_key": "qwen3.5-2b",
     "maker": "Alibaba",
     "kind": "chat",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 59.095063985374765,
     "raw": {
      "metric": "accuracy",
      "value": 0.6727605118829981,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1215"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision Qwen/Qwen3.5-2B@15852e8), one question per request, plain answer line CEVAP: X (no answer schema); sampling min_p 0.0, presence_penalty 1.5, repetition_penalty 1.0, seed 0, temperature 1.0, top_k 20, top_p 0.95, thinking off; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "LiquidAI/d1-2B",
     "commit": "10e7671",
     "name": "d1-2B",
     "tag": "",
     "model_key": "liquid-d1-2b",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": 1.92,
     "params_active": 1.92,
     "quantization": "BF16",
     "score_0_100": 57.03839122486289,
     "raw": {
      "metric": "accuracy",
      "value": 0.6563071297989032,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.07860799520284192,
     "improvement_points": null,
     "runs": [
      "english:708"
     ],
     "notes": "d1 server (Liquid's model code in Transformers, the model card's call), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiDE-0.8B",
     "commit": "9eacdee",
     "name": "GLiDE 0.8B",
     "tag": "",
     "model_key": "glide-0.8b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.75,
     "params_active": 0.75,
     "quantization": "BF16",
     "score_0_100": 50.86837294332723,
     "raw": {
      "metric": "accuracy",
      "value": 0.6069469835466179,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.12502588537695936,
     "improvement_points": null,
     "runs": [
      "english:925"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-0.8B-Instant",
     "commit": "6496ae641cc71b149ce9dad69bd97f2a44e7bf11",
     "name": "StartLux-Decision-v1.1-0.8B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-0.8b-instant",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 49.49725776965264,
     "raw": {
      "metric": "accuracy",
      "value": 0.5959780621572212,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.0874876350924171,
     "improvement_points": null,
     "runs": [
      "english:1011"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "OmniJev/OneJev1.6-0.8B",
     "commit": "bcd88a382f314b26cb7b1b7f369c1ed3502ee7a0",
     "name": "OneJev1.6-0.8B",
     "tag": "",
     "model_key": "onejev1.6-0.8b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 0.85,
     "params_active": 0.85,
     "quantization": "BF16",
     "score_0_100": 49.26873857404021,
     "raw": {
      "metric": "accuracy",
      "value": 0.5941499085923218,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03740665996343693,
     "improvement_points": null,
     "runs": [
      "english:1025"
     ],
     "notes": "Flide server (flide 0.1.11, the card's server for OneJev 1.6), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiDE-0.4B",
     "commit": "6bda7ab",
     "name": "GLiDE 0.4B",
     "tag": "",
     "model_key": "glide-0.4b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.4,
     "params_active": 0.4,
     "quantization": "BF16",
     "score_0_100": 49.26873857404021,
     "raw": {
      "metric": "accuracy",
      "value": 0.5941499085923218,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.047887521004017934,
     "improvement_points": null,
     "runs": [
      "english:933"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the kept reading questions in English"
    },
    {
     "repo": "PelaAI/KnowLine-0.8B-Gen1",
     "commit": "6b27831f7b783c03278bd946d4d8f4f7a98a3798",
     "name": "KnowLine-0.8B-Gen1",
     "tag": "",
     "model_key": "knowline-0.8b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 48.58318098720292,
     "raw": {
      "metric": "accuracy",
      "value": 0.5886654478976234,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.08129406814013407,
     "improvement_points": null,
     "runs": [
      "english:1001"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": "knowledgator/gliclass-v4-edge",
     "commit": "885af1af8a07edf4ed34df83ca0d995ed8201283",
     "name": "GLiClass v4 Edge",
     "tag": "",
     "model_key": "gliclass-v4-edge",
     "maker": "Knowledgator",
     "kind": "decision",
     "params_total": 0.44,
     "params_active": 0.44,
     "quantization": "BF16",
     "score_0_100": 47.89762340036562,
     "raw": {
      "metric": "accuracy",
      "value": 0.583180987202925,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.03781915740672614,
     "improvement_points": null,
     "runs": [
      "english:1042"
     ],
     "notes": "GLiClass server (our servers/gliclass_server.py: the card's gliclass pipeline and its System One mapping), self-hosted, BF16; the kept reading questions in English"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Decider 1",
     "tag": "",
     "model_key": "meragpt-decider-1",
     "maker": "meraGPT",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 44.698354661791576,
     "raw": {
      "metric": "accuracy",
      "value": 0.5575868372943327,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.05338702010968936,
     "improvement_points": null,
     "runs": [
      "english:596"
     ],
     "notes": "meraGPT System One API; the kept reading questions in English"
    },
    {
     "repo": "KRAFTON/Raon-1-nano",
     "commit": "fcf2d1d",
     "name": "Raon-1 nano",
     "tag": "reasoning off · chat",
     "model_key": "raon-1-nano",
     "maker": "KRAFTON",
     "kind": "chat",
     "params_total": 3.53,
     "params_active": 3.53,
     "quantization": "BF16",
     "score_0_100": 42.18464351005484,
     "raw": {
      "metric": "accuracy",
      "value": 0.5374771480804388,
      "questions": 547,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "english:1152"
     ],
     "notes": "our GPU (logic's server, KRAFTON's vLLM fork (krafton-ai/vllm@13a5dd8; not a vLLM release) BF16, revision KRAFTON/Raon-1-nano@fcf2d1d), one question per request, plain answer line ANSWER: X (no answer schema); sampling presence_penalty 1.5, seed 0, temperature 0.1, thinking off; output cap 32,768; the kept reading questions in English"
    },
    {
     "repo": "StrandsAgents/strands-decider-2B-hobson-v21",
     "commit": "2b52a62",
     "name": "Strands Decider 2B",
     "tag": "",
     "model_key": "strands-decider-2b",
     "maker": "Strands Agents",
     "kind": "decision",
     "params_total": 2.29,
     "params_active": 2.29,
     "quantization": "BF16",
     "score_0_100": 31.901279707495426,
     "raw": {
      "metric": "accuracy",
      "value": 0.45521023765996343,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.10518775137111518,
     "improvement_points": null,
     "runs": [
      "english:702"
     ],
     "notes": "Strands Decider server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya",
     "tag": "",
     "model_key": "laya",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.42,
     "params_active": 0.42,
     "quantization": null,
     "score_0_100": 3.5648994515539294,
     "raw": {
      "metric": "accuracy",
      "value": 0.22851919561243145,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.12050274223034743,
     "improvement_points": null,
     "runs": [
      "english:726"
     ],
     "notes": "Laya server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "fastino/GLiNER2.5-multi-Decide",
     "commit": "a35a0cd",
     "name": "GLiNER2.5-multi-Decide",
     "tag": "",
     "model_key": "gliner2.5-multi-decide",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": 1.5082266910420445,
     "raw": {
      "metric": "accuracy",
      "value": 0.21206581352833637,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.2718228085930212,
     "improvement_points": null,
     "runs": [
      "english:714"
     ],
     "notes": "GLiNER2 server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "fastino/gliner2.5-multi-v1",
     "commit": "2ca71aa",
     "name": "GLiNER2.5 Multi",
     "tag": "",
     "model_key": "gliner2.5-multi-base",
     "maker": "Fastino",
     "kind": "chat",
     "params_total": 0.29,
     "params_active": 0.29,
     "quantization": "FP16",
     "score_0_100": -4.661791590493604,
     "raw": {
      "metric": "accuracy",
      "value": 0.16270566727605118,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.36424333126823905,
     "improvement_points": null,
     "runs": [
      "english:720"
     ],
     "notes": "GLiNER2 server, self-hosted; the kept reading questions in English"
    },
    {
     "repo": "convaiinnovations/laya",
     "commit": "c5d7873",
     "name": "Laya Multilingual",
     "tag": "",
     "model_key": "laya-multilingual",
     "maker": "Convai Innovations",
     "kind": "decision",
     "params_total": 0.32,
     "params_active": 0.32,
     "quantization": null,
     "score_0_100": -5.804387568555759,
     "raw": {
      "metric": "accuracy",
      "value": 0.15356489945155394,
      "questions": 547,
      "options": 5
     },
     "calibration_error": 0.4436060329067641,
     "improvement_points": null,
     "runs": [
      "english:732"
     ],
     "notes": "Laya server, self-hosted; the kept reading questions in English"
    }
   ]
  },
  {
   "bench": "reading-vision",
   "title": "Reading · Vision",
   "url": "https://zekiworks.github.io/analysis/decision-models/vision/",
   "updated": "2026-11-16T20:35:27Z",
   "score": {
    "zero": "chance: 1 of each question's options right on every question (accuracy 20% with 5 options)",
    "hundred": "every one of the 763 questions that depend on the printed page right, read from the page image alone (accuracy 100%)"
   },
   "models": [
    {
     "repo": null,
     "commit": null,
     "name": "Muse Ember",
     "tag": "max · chat",
     "model_key": "muse-ember",
     "maker": "Meta",
     "kind": "chat",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 82.79816513761469,
     "raw": {
      "metric": "accuracy",
      "value": 0.8623853211009175,
      "questions": 763,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "vision:128",
      "vision:129",
      "vision:130",
      "vision:131",
      "vision:132",
      "vision:133",
      "vision:134",
      "vision:135",
      "vision:136",
      "vision:137",
      "vision:138",
      "vision:139",
      "vision:140",
      "vision:141",
      "vision:142",
      "vision:143",
      "vision:144",
      "vision:145",
      "vision:146",
      "vision:147",
      "vision:148",
      "vision:149",
      "vision:150",
      "vision:151",
      "vision:152",
      "vision:153",
      "vision:154",
      "vision:155",
      "vision:156",
      "vision:157",
      "vision:158",
      "vision:159",
      "vision:160",
      "vision:161",
      "vision:162",
      "vision:163",
      "vision:164",
      "vision:165",
      "vision:166",
      "vision:167",
      "vision:168",
      "vision:169"
     ],
     "notes": "OpenRouter chat API, Meta's own endpoint, fallbacks off, output cap 65,536 tokens, reasoning max; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Gemini Decision 1",
     "tag": "20261111",
     "model_key": "gemini-decision-1",
     "maker": "Google",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 58.0602883355177,
     "raw": {
      "metric": "accuracy",
      "value": 0.6644823066841415,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.05173363518815327,
     "improvement_points": null,
     "runs": [
      "vision:116"
     ],
     "notes": "OpenRouter decisions API, Google AI Studio endpoint; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1-ultra",
     "tag": "",
     "model_key": "liquid-d1-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 37.5819134993447,
     "raw": {
      "metric": "accuracy",
      "value": 0.5006553079947575,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.034957632368552746,
     "improvement_points": null,
     "runs": [
      "vision:2",
      "vision:5"
     ],
     "notes": "Liquid AI decisions API; the page image only (condition B)"
    },
    {
     "repo": "fastino/GLiDE-27B",
     "commit": "4d32743",
     "name": "GLiDE 27B",
     "tag": "",
     "model_key": "glide-27b",
     "maker": "Fastino",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 37.41808650065531,
     "raw": {
      "metric": "accuracy",
      "value": 0.49934469200524245,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.04281848707465709,
     "improvement_points": null,
     "runs": [
      "vision:68"
     ],
     "notes": "GLiDE server (Fastino's release code), self-hosted; the page image only (condition B)"
    },
    {
     "repo": "PelaAI/KnowLine-31B-Gen1.5",
     "commit": "8e56aa1b236a6dd1064dcfd22ba1e5e360e4c140",
     "name": "KnowLine-31B-Gen1.5",
     "tag": "",
     "model_key": "knowline-31b-gen1.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 31.27,
     "params_active": 31.27,
     "quantization": "BF16",
     "score_0_100": 37.254259501965926,
     "raw": {
      "metric": "accuracy",
      "value": 0.4980340760157274,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.06146547860784749,
     "improvement_points": null,
     "runs": [
      "vision:86"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": "vllm-sr/d3",
     "commit": "20f37b9",
     "name": "d3",
     "tag": "",
     "model_key": "vllm-sr-d3",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 35.61598951507208,
     "raw": {
      "metric": "accuracy",
      "value": 0.4849279161205767,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.03745645138174668,
     "improvement_points": null,
     "runs": [
      "vision:41"
     ],
     "notes": "vLLM System One endpoint (nightly), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2",
     "tag": "",
     "model_key": "liquid-d1.2",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 33.65006553079948,
     "raw": {
      "metric": "accuracy",
      "value": 0.4692005242463958,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.05624499494183653,
     "improvement_points": null,
     "runs": [
      "vision:571"
     ],
     "notes": "Liquid AI decisions API; the page image only (condition B)"
    },
    {
     "repo": "nace-ai/drex-max-v1",
     "commit": "cac65e1",
     "name": "Drex Max",
     "tag": "",
     "model_key": "drex-max-v1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 33.48623853211009,
     "raw": {
      "metric": "accuracy",
      "value": 0.46788990825688076,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.09308429546459711,
     "improvement_points": null,
     "runs": [
      "vision:56"
     ],
     "notes": "Drex Max server (Nace's loader, drex_max), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "SubQ Snap 1.1",
     "tag": "",
     "model_key": "subq-snap-1.1",
     "maker": "Subquadratic",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 32.17562254259502,
     "raw": {
      "metric": "accuracy",
      "value": 0.45740498034076016,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.11736391430054234,
     "improvement_points": null,
     "runs": [
      "vision:62"
     ],
     "notes": "SubQ Decisions API (public preview); the page image only (condition B)"
    },
    {
     "repo": "vllm-sr/d3-ultra",
     "commit": "52ba9e7",
     "name": "d3-ultra",
     "tag": "",
     "model_key": "vllm-sr-d3-ultra",
     "maker": "vLLM Semantic Router",
     "kind": "decision",
     "params_total": 26.09,
     "params_active": 26.09,
     "quantization": "BF16",
     "score_0_100": 31.847968545216247,
     "raw": {
      "metric": "accuracy",
      "value": 0.45478374836173,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.18775177910999716,
     "improvement_points": null,
     "runs": [
      "vision:29"
     ],
     "notes": "d3 package server (transformers, the release code), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "d1.2-ultra",
     "tag": "",
     "model_key": "liquid-d1.2-ultra",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 31.684141546526863,
     "raw": {
      "metric": "accuracy",
      "value": 0.4534731323722149,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.028242830658147308,
     "improvement_points": null,
     "runs": [
      "vision:570"
     ],
     "notes": "Liquid AI decisions API; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Drex Max v1.1",
     "tag": "",
     "model_key": "drex-max-1.1",
     "maker": "Nace.AI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 31.356487549148106,
     "raw": {
      "metric": "accuracy",
      "value": 0.4508519003931848,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.05152597785022986,
     "improvement_points": null,
     "runs": [
      "vision:58"
     ],
     "notes": "Nace.AI System One API; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "Jev 1.14",
     "tag": "",
     "model_key": "jev-1.14.0",
     "maker": "TypeSafe",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 29.062909567496725,
     "raw": {
      "metric": "accuracy",
      "value": 0.4325032765399738,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.03672346002621207,
     "improvement_points": null,
     "runs": [
      "vision:1",
      "vision:6"
     ],
     "notes": "TypeSafe System One API; the page image only (condition B)"
    },
    {
     "repo": "google/gemma-4-E4B-it",
     "commit": "4ecc2ba",
     "name": "Gemma 4 E4B",
     "tag": "default reasoning · chat",
     "model_key": "gemma-4-e4b",
     "maker": "Google",
     "kind": "chat",
     "params_total": 8.0,
     "params_active": 8.0,
     "quantization": "BF16",
     "score_0_100": 28.57142857142857,
     "raw": {
      "metric": "accuracy",
      "value": 0.42857142857142855,
      "questions": 763,
      "options": 5
     },
     "calibration_error": null,
     "improvement_points": null,
     "runs": [
      "vision:465"
     ],
     "notes": "our GPU (logic's server, vLLM v0.30.0 BF16, revision google/gemma-4-E4B-it@4ecc2ba), one question per request, plain answer line CEVAP: X (no answer schema); sampling temperature 0.0; output cap 8,192; the page image only (condition B)"
    },
    {
     "repo": "OmniJev/OneJev-27B",
     "commit": "a5e5f55",
     "name": "OneJev-27B",
     "tag": "",
     "model_key": "onejev-27b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 27.36,
     "params_active": 27.36,
     "quantization": "BF16",
     "score_0_100": 26.605504587155966,
     "raw": {
      "metric": "accuracy",
      "value": 0.41284403669724773,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.04855050196592398,
     "improvement_points": null,
     "runs": [
      "vision:28",
      "vision:35"
     ],
     "notes": "OneJev server (qev, OmniJev's release code), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": null,
     "commit": null,
     "name": "GPT-6 Luna",
     "tag": "Decisions API",
     "model_key": "gpt-6-luna",
     "maker": "OpenAI",
     "kind": "decision",
     "params_total": null,
     "params_active": null,
     "quantization": null,
     "score_0_100": 15.792922673656621,
     "raw": {
      "metric": "accuracy",
      "value": 0.32634338138925295,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.05684141546526838,
     "improvement_points": null,
     "runs": [
      "vision:3",
      "vision:4"
     ],
     "notes": "OpenAI Decisions API (public beta); the page image only (condition B)"
    },
    {
     "repo": "PelaAI/KnowLine-4B-Gen5.5",
     "commit": "779944d2cc8ffa19cef386aee83c83bcb967ade2",
     "name": "KnowLine 4B Gen5.5",
     "tag": "",
     "model_key": "knowline-4b-gen5.5",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 4.66,
     "params_active": 4.66,
     "quantization": "FP8 at load",
     "score_0_100": 12.680209698558324,
     "raw": {
      "metric": "accuracy",
      "value": 0.30144167758846657,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.11030108065283414,
     "improvement_points": null,
     "runs": [
      "vision:92"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, FP8 at load; the page image only (condition B)"
    },
    {
     "repo": "OmniJev/OneJev1.6-0.8B",
     "commit": "bcd88a382f314b26cb7b1b7f369c1ed3502ee7a0",
     "name": "OneJev1.6-0.8B",
     "tag": "",
     "model_key": "onejev1.6-0.8b",
     "maker": "OmniJev",
     "kind": "decision",
     "params_total": 0.85,
     "params_active": 0.85,
     "quantization": "BF16",
     "score_0_100": 8.09305373525557,
     "raw": {
      "metric": "accuracy",
      "value": 0.26474442988204455,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.04235339580602877,
     "improvement_points": null,
     "runs": [
      "vision:122"
     ],
     "notes": "Flide server (flide 0.1.11, the card's server for OneJev 1.6), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-0.8B-Instant",
     "commit": "6496ae641cc71b149ce9dad69bd97f2a44e7bf11",
     "name": "StartLux-Decision-v1.1-0.8B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-0.8b-instant",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 5.307994757536044,
     "raw": {
      "metric": "accuracy",
      "value": 0.24246395806028834,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.01694368756323348,
     "improvement_points": null,
     "runs": [
      "vision:110"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": "PelaAI/KnowLine-0.8B-Gen1",
     "commit": "6b27831f7b783c03278bd946d4d8f4f7a98a3798",
     "name": "KnowLine-0.8B-Gen1",
     "tag": "",
     "model_key": "knowline-0.8b-gen1",
     "maker": "PelaAI",
     "kind": "decision",
     "params_total": 0.87,
     "params_active": 0.87,
     "quantization": "BF16",
     "score_0_100": 4.816513761467894,
     "raw": {
      "metric": "accuracy",
      "value": 0.23853211009174313,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.014523651044381,
     "improvement_points": null,
     "runs": [
      "vision:104"
     ],
     "notes": "KnowLine server (knowline_server.py over SGLang 0.5.21, PelaAI's release code), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": "startlux-models/StartLux-Decision-v1.1-2B-Instant",
     "commit": "f1e6cd147dda82c9ffa3e8e725d5c1f175566fb1",
     "name": "StartLux-Decision-v1.1-2B-Instant",
     "tag": "",
     "model_key": "startlux-decision-v1.1-2b",
     "maker": "StartLux",
     "kind": "decision",
     "params_total": 2.27,
     "params_active": 2.27,
     "quantization": "BF16",
     "score_0_100": 4.325032765399739,
     "raw": {
      "metric": "accuracy",
      "value": 0.2346002621231979,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.036409141711210374,
     "improvement_points": null,
     "runs": [
      "vision:98"
     ],
     "notes": "StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16; the page image only (condition B)"
    },
    {
     "repo": "LiquidAI/d1-2B",
     "commit": "10e7671",
     "name": "d1-2B",
     "tag": "",
     "model_key": "liquid-d1-2b",
     "maker": "Liquid AI",
     "kind": "decision",
     "params_total": 1.92,
     "params_active": 1.92,
     "quantization": "BF16",
     "score_0_100": 4.161205766710356,
     "raw": {
      "metric": "accuracy",
      "value": 0.23328964613368283,
      "questions": 763,
      "options": 5
     },
     "calibration_error": 0.010668762025147863,
     "improvement_points": null,
     "runs": [
      "vision:60"
     ],
     "notes": "d1 server (Liquid's model code in Transformers, the model card's call), self-hosted, BF16; the page image only (condition B)"
    }
   ]
  }
 ]
}
