{
 "setups": [
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Perplexity Decider 27B v1.1 (local GPU)",
   "model_id": "pplx-decider-v1.1-27b",
   "model_type": "decision model",
   "name": "Perplexity Decider 27B v1.1 (local GPU)",
   "notes": "Perplexity's released checkpoint on our GPUs; moved to a DGX Spark on 16 Oct 11:04 UTC (same checkpoint and server arguments; 20 replayed requests: same play 20/20)",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16; NVIDIA DGX Spark (GB10) from 16 Oct",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "decider-v1-1-27b",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights",
    "price.date"
   ],
   "model": "SubQ Snap Preview",
   "model_id": "subq-snap-preview",
   "model_type": "decision model",
   "name": "SubQ Snap Preview",
   "notes": null,
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 0.05,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "SubQ API",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "subq-snap-preview",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Clef",
   "model_id": "clef",
   "model_type": "decision model",
   "name": "Clef",
   "notes": "moved to a DGX Spark on 16 Oct 11:04 UTC (same checkpoint and server arguments; 20 replayed requests: same play 20/20)",
   "option_groups": "up to 110 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16; NVIDIA DGX Spark (GB10) from 16 Oct",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "clef",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Open-Jev 27B v1.1",
   "model_id": null,
   "model_type": "decision model",
   "name": "Open-Jev 27B v1.1",
   "notes": "its server takes no model field; stayed on the same GPU throughout",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "open-jev-27b-v1-1",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "sampling"
   ],
   "model": "Cygnet",
   "model_id": "cygnet",
   "model_type": "decision model",
   "name": "Cygnet",
   "notes": "Google's gemma-4-12B-it behind a server that labels the options with letters and reads the letter probabilities; moved to a DGX Spark on 16 Oct 11:04 UTC (20 replayed requests: same play 20/20)",
   "option_groups": "up to 26 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16; NVIDIA DGX Spark (GB10) from 16 Oct",
   "reasoning": "none (letter read directly)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": null,
   "setting": null,
   "slug": "cygnet-gemma-4-12b",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Random",
   "model_id": null,
   "model_type": "program",
   "name": "Random",
   "notes": "random legal play",
   "option_groups": null,
   "price": null,
   "provider": "program: random legal play",
   "reasoning": null,
   "rules_text": null,
   "runs_on": "CPU",
   "sampling": null,
   "setting": null,
   "slug": "random",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "tools/v3/page_v3.py"
   ],
   "weights": null
  },
  {
   "confidence_source": "none",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Safest",
   "model_id": null,
   "model_type": "program",
   "name": "Safest",
   "notes": "the play the fewest opponent rolls hit",
   "option_groups": null,
   "price": null,
   "provider": "program: fewest hitting rolls",
   "reasoning": null,
   "rules_text": null,
   "runs_on": "CPU",
   "sampling": null,
   "setting": null,
   "slug": "safest",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "tools/v3/page_v3.py"
   ],
   "weights": null
  },
  {
   "confidence_source": "none",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "GNUbg weak",
   "model_id": null,
   "model_type": "program",
   "name": "GNUbg weak",
   "notes": "GNU Backgammon 1.08, Beginner level (0-ply, deterministic noise 0.060); fixed anchor in every league",
   "option_groups": null,
   "price": null,
   "provider": "GNU Backgammon 1.08, Beginner level (0-ply, noise 0.060), x86-64 build (see About the anchor)",
   "reasoning": null,
   "rules_text": null,
   "runs_on": "CPU",
   "sampling": null,
   "setting": null,
   "slug": "gnubg-weak",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "tools/v3/page_v3.py"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights"
   ],
   "model": "Liquid d1-ultra",
   "model_id": "d1-ultra",
   "model_type": "decision model",
   "name": "Liquid d1-ultra",
   "notes": "Liquid AI decisions API, paid tier",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-15",
    "input_per_mtok": 0.042,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "Liquid AI decisions API",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "liquid-d1-ultra",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights"
   ],
   "model": "Jev 1.14.0",
   "model_id": "jev-1.14.0",
   "model_type": "decision model",
   "name": "Jev 1.14.0",
   "notes": "TypeSafe System One API",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-16",
    "input_per_mtok": 0.042,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "TypeSafe API",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "jev-1-14-0",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights"
   ],
   "model": "Jev 1.13.0",
   "model_id": "jev-1.13.0",
   "model_type": "decision model",
   "name": "Jev 1.13.0",
   "notes": "TypeSafe System One API",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-08",
    "input_per_mtok": 0.042,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "TypeSafe API",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "jev-1-13-0",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights"
   ],
   "model": "Celeris-1 Decision",
   "model_id": "celeris-1-decision",
   "model_type": "decision model",
   "name": "Celeris-1 Decision",
   "notes": "probabilities come rounded to two decimals",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-10",
    "input_per_mtok": 0.04,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "Celeris API (probabilities rounded to 2 decimals)",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "celeris-1-decision",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights"
   ],
   "model": "GPT-6 Luna (Decisions API)",
   "model_id": "gpt-6-luna",
   "model_type": "decision model",
   "name": "GPT-6 Luna (Decisions API)",
   "notes": "OpenAI Decisions API (public beta); an earlier League 1 entry through the ChatGPT subscription (chat model) was removed on 19 Oct",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-08",
    "input_per_mtok": 0.1,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "OpenAI Decisions API",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "gpt-6-luna",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Perplexity Decider Flash v1.2",
   "model_id": "pplx-decider-flash-v1.2",
   "model_type": "decision model",
   "name": "Perplexity Decider Flash v1.2",
   "notes": "Perplexity's pplx-decider-flash-v1.2 (from Qwen3.5-4B) through the pplx_decider package its checkpoint ships, one request at a time; price is Perplexity's list price for the same model",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "decider-flash-v1-2",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Winnow-12B",
   "model_id": "winnow-12b",
   "model_type": "decision model",
   "name": "Winnow-12B",
   "notes": "a Gemma 4 12B IT fine-tune, run through its author's inference server with the author's configuration (Q8_0 GGUF); no published price",
   "option_groups": "up to 40 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), Q8_0 GGUF, author's server",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "winnow-12b",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights"
   ],
   "model": "Inception Mercury Decide",
   "model_id": "inception/mercury-decide",
   "model_type": "decision model",
   "name": "Inception Mercury Decide",
   "notes": "Inception's Mercury Decide through OpenRouter's decision route, paid listing; latency includes OpenRouter's own time",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-17",
    "input_per_mtok": 0.02,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "OpenRouter (Inception)",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "mercury-decide",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "weights",
    "price"
   ],
   "model": "Fastino GLiDE",
   "model_id": "fastino/GLiDE",
   "model_type": "decision model",
   "name": "Fastino GLiDE",
   "notes": "Fastino's hosted decision API; every stored response (all its League 1 matches, 20 Oct) names the model glide_v2, as did the API on 22 Oct",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-11-10",
    "input_per_mtok": 0.04,
    "label": "API price",
    "output_per_mtok": 0.0,
    "source": "https://docs.fastino.ai/pricing (list price × our recorded tokens; the turns carry no cost)"
   },
   "provider": "Fastino API (responses name the model glide_v2)",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "fastino-glide",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [],
   "model": "Perplexity Decider 27B v1.2 (local GPU)",
   "model_id": "pplx-decider-v1.2",
   "model_type": "decision model",
   "name": "Perplexity Decider 27B v1.2 (local GPU)",
   "notes": "Perplexity's pplx-decider-v1.2 through the pplx_decider package its checkpoint ships, one request at a time; joined late (20 Oct); price is Perplexity's list price for the same model",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "decider-v1-2-27b",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "league-1",
   "date_run": null,
   "missing": [
    "price"
   ],
   "model": "Fastino GLiDE v1.2",
   "model_id": "fastino/GLiDE @ a7e45896 (glide-v1.2)",
   "model_type": "decision model",
   "name": "Fastino GLiDE v1.2",
   "notes": "Fastino's GLiDE v1.2 weights (Hugging Face fastino/GLiDE, Fastino Community License) through the glide/server.py its repository ships, bf16, one request per forward pass; joined late (22 Oct)",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16, Fastino's released weights and server",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "fastino-glide-v1-2-local",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league1.json",
    "freeze/league1-additions.json",
    "tools/v3/page_v3.py",
    "src/players/decision.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Gemini 3.8 Flash",
   "model_id": "gemini-3.8-flash-low",
   "model_type": "chat model, one letter",
   "name": "Gemini 3.8 Flash (low)",
   "notes": "Google Antigravity subscription from 19 Oct 12:46 UTC (Antigravity model id gemini-3.8-flash-low, model version gemini-3.8-flash); before that it waited for a Gemini API key; answer letter constrained to the option letters",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 0.75,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 3.75
   },
   "provider": "Google Antigravity subscription",
   "reasoning": "thinking level low",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "gemini-3-8-flash-low",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Gemini 3.8 Flash",
   "model_id": "gemini-3.8-flash-high",
   "model_type": "chat model, one letter",
   "name": "Gemini 3.8 Flash (high)",
   "notes": "Google Antigravity subscription from 19 Oct 12:46 UTC (Antigravity model id gemini-3.8-flash-high, model version gemini-3.8-flash); before that it waited for a Gemini API key; answer letter constrained to the option letters",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 0.75,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 3.75
   },
   "provider": "Google Antigravity subscription",
   "reasoning": "thinking level high",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "temperature 1.0",
   "setting": "high",
   "slug": "gemini-3-8-flash-high",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Claude Sonnet 5.5",
   "model_id": "claude-sonnet-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Sonnet 5.5 (low)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 10.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort low",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "claude-sonnet-5-5-low",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Claude Sonnet 5.5",
   "model_id": "claude-sonnet-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Sonnet 5.5 (high)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 10.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort high",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "claude-sonnet-5-5-high",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "GPT-6.1 Sol",
   "model_id": "gpt-6.1-sol",
   "model_type": "chat model, one letter",
   "name": "GPT-6.1 Sol (low)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 10.0
   },
   "provider": "ChatGPT subscription",
   "reasoning": "reasoning effort low",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "gpt-6-1-sol-low",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "GPT-6.1 Sol",
   "model_id": "gpt-6.1-sol",
   "model_type": "chat model, one letter",
   "name": "GPT-6.1 Sol (high)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 10.0
   },
   "provider": "ChatGPT subscription",
   "reasoning": "reasoning effort high",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "gpt-6-1-sol-high",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "DeepSeek-V4.1-Flash",
   "model_id": "deepseek-v4.1-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "DeepSeek-V4.1-Flash (off)",
   "notes": "Fireworks via OpenRouter (model ID deepseek/deepseek-v4.1-flash, quantization not published, at most 5 top logprobs) until 20 Oct 21:56 UTC, then the published weights on our GPUs (SGLang), still 5 top logprobs; the price stays Fireworks' list price for comparison",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 0.3,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": 1.2
   },
   "provider": "Fireworks via OpenRouter (quantization not published) until 20 Oct 21:56 UTC, then 4 × NVIDIA RTX PRO 6000 Blackwell (96 GB) (SGLang, published FP4/FP8 weights, 17 Oct revision)",
   "reasoning": "thinking off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "deepseek-v4-1-flash-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "aliases": [
    "DeepSeek V4.1 Flash (on)"
   ],
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "DeepSeek-V4.1-Flash",
   "model_id": "deepseek-v4.1-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "DeepSeek-V4.1-Flash (low)",
   "notes": "Fireworks via OpenRouter (model ID deepseek/deepseek-v4.1-flash, quantization not published) until 20 Oct 21:56 UTC, then the published weights on our GPUs (SGLang); letter probabilities read from the answer after the reasoning; the price stays Fireworks' list price for comparison",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 0.3,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": 1.2
   },
   "provider": "Fireworks via OpenRouter (quantization not published) until 20 Oct 21:56 UTC, then 4 × NVIDIA RTX PRO 6000 Blackwell (96 GB) (SGLang, published FP4/FP8 weights, 17 Oct revision)",
   "reasoning": "thinking on, reasoning effort low",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "deepseek-v4-1-flash-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "aliases": [
    "GLM-5.3 (on)"
   ],
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "GLM-5.3",
   "model_id": "z-ai/glm-5.3",
   "model_type": "chat model, answer-letter probabilities",
   "name": "GLM-5.3 (low)",
   "notes": "OpenRouter, pinned to Fireworks (no fallbacks); quantization not published",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 1.4,
    "label": "API price",
    "output_per_mtok": 4.4
   },
   "provider": "Fireworks via OpenRouter, quantization not published",
   "reasoning": "reasoning effort low (reasoning is mandatory on this endpoint, so no off setup)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "glm-5-3-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "GNUbg weak",
   "model_id": null,
   "model_type": "program",
   "name": "GNUbg weak",
   "notes": "GNU Backgammon 1.08, Beginner level (0-ply, deterministic noise 0.060); fixed anchor in every league",
   "option_groups": null,
   "price": null,
   "provider": "GNU Backgammon 1.08, Beginner level (0-ply, noise 0.060), x86-64 build (see About the anchor)",
   "reasoning": null,
   "rules_text": null,
   "runs_on": "CPU",
   "sampling": null,
   "setting": null,
   "slug": "gnubg-weak",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "tools/v3/page_v3.py"
   ],
   "weights": null
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [
    "weights",
    "price.date"
   ],
   "model": "Qwen3.8 Flash",
   "model_id": "qwen/qwen3.8-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Qwen3.8 Flash (off)",
   "notes": "OpenRouter, pinned to Alibaba (its only provider); replaces Qwen3.8 Flash Next, which is not on OpenRouter; quantization not published",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 0.15,
    "label": "API price",
    "output_per_mtok": 0.47
   },
   "provider": "Alibaba via OpenRouter, quantization not published",
   "reasoning": "reasoning off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "qwen3-8-flash-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json"
   ],
   "weights": null
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "LFM2.5-24B-A2B",
   "model_id": "lfm2.5-24b-a2b",
   "model_type": "chat model, answer-letter probabilities",
   "name": "LFM2.5-24B-A2B (off)",
   "notes": "Liquid AI's own API until its matches were voided on 19 Oct 22:20 UTC; from 22:26 UTC Liquid's bf16 checkpoint on our GPU (vLLM)",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16 (vLLM); NVIDIA DGX Spark (GB10) from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)",
   "reasoning": "no reasoning (the letter continues a prefilled empty thinking block and Answer:)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "lfm2-5-24b-a2b-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "LFM2.5-24B-A2B",
   "model_id": "lfm2.5-24b-a2b",
   "model_type": "chat model, answer-letter probabilities",
   "name": "LFM2.5-24B-A2B (on)",
   "notes": "Liquid AI's own API until its matches were voided on 19 Oct 22:20 UTC; from 22:26 UTC Liquid's bf16 checkpoint on our GPU (vLLM); letter read from a prefilled Answer: after the reasoning",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16 (vLLM); first decisions on Liquid AI's API (voided); NVIDIA DGX Spark (GB10) from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)",
   "reasoning": "reasons on every answer (its template has no switch or effort setting; \"low\" is sent and ignored)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "on",
   "slug": "lfm2-5-24b-a2b-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "Gemma 4 31B",
   "model_id": "gemma-4-31B-it",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Gemma 4 31B (on)",
   "notes": "moved between our GPUs and a DGX Spark during the league with the same image, checkpoint and arguments",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), fp8 (vLLM); NVIDIA DGX Spark (GB10) 19 Oct 18:36-22:26 UTC and from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)",
   "reasoning": "thinking on (template default; no effort setting)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "on",
   "slug": "gemma-4-31b-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "Gemma 4 31B",
   "model_id": "gemma-4-31B-it",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Gemma 4 31B (off)",
   "notes": "moved between our GPUs and a DGX Spark during the league with the same image, checkpoint and arguments",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), fp8 (vLLM); NVIDIA DGX Spark (GB10) 19 Oct 18:36-22:26 UTC and from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)",
   "reasoning": "thinking off",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "gemma-4-31b-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "Qwen3.8-27B",
   "model_id": "qwen3.8-27b",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Qwen3.8-27B (off)",
   "notes": "Cerebras via OpenRouter (model ID qwen/qwen3.8-27b, fp16) before 19 Oct 22:26 UTC, where it finished no match; then on our GPU (vLLM, fp16); unconstrained letter",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), fp16 (vLLM); first decisions on Cerebras via OpenRouter; NVIDIA DGX Spark (GB10) from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)",
   "reasoning": "thinking off",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "qwen3-8-27b-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "aliases": [
    "Qwen3.8-27B (on)"
   ],
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": null,
   "missing": [],
   "model": "Qwen3.8-27B",
   "model_id": "qwen3.8-27b",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Qwen3.8-27B (low)",
   "notes": "Cerebras via OpenRouter (model ID qwen/qwen3.8-27b, fp16) before 19 Oct 22:26 UTC, where it finished no match (its first 4 decisions came from there); then on our GPU (vLLM, fp16); letter read from a prefilled Answer: after the reasoning",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), fp16 (vLLM); first decisions on Cerebras via OpenRouter; NVIDIA DGX Spark (GB10) from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)",
   "reasoning": "thinking on, reasoning effort low (the chat template's reasoning_effort)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "qwen3-8-27b-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league2.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "freeze/league2-changes.json",
    "freeze/league2-additions.json",
    "leagues/league2-start.json",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "aliases": [
    "GPT-6 Astra (off)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "GPT-6 Astra",
   "model_id": "gpt-6-astra",
   "model_type": "chat model, one letter",
   "name": "GPT-6 Astra (low)",
   "notes": "ChatGPT subscription (Codex backend)",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 10.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50.0
   },
   "provider": "ChatGPT subscription",
   "reasoning": "reasoning effort low, the lowest gpt-6-astra accepts (none and minimal are rejected)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "gpt-6-astra-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "GPT-6 Astra (on)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "GPT-6 Astra",
   "model_id": "gpt-6-astra",
   "model_type": "chat model, one letter",
   "name": "GPT-6 Astra (medium)",
   "notes": "ChatGPT subscription (Codex backend)",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 10.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50.0
   },
   "provider": "ChatGPT subscription",
   "reasoning": "reasoning effort medium, the model's default",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "medium",
   "slug": "gpt-6-astra-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Claude Opus 5.5 (off)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Claude Opus 5.5",
   "model_id": "claude-opus-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Opus 5.5 (low)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 4.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 20.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort low, the lowest Opus 5.5 accepts (adaptive thinking only)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "claude-opus-5-5-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Claude Opus 5.5 (on)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Claude Opus 5.5",
   "model_id": "claude-opus-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Opus 5.5 (high)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 4.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 20.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort high, the default (adaptive thinking)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "claude-opus-5-5-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Claude Fable 5.1 (off)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Claude Fable 5.1",
   "model_id": "claude-fable-5-1",
   "model_type": "chat model, one letter",
   "name": "Claude Fable 5.1 (low)",
   "notes": "the newest Fable the subscription serves",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 10.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort low, the lowest it accepts (adaptive thinking only)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "claude-fable-5-1-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Claude Fable 5.1 (on)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Claude Fable 5.1",
   "model_id": "claude-fable-5-1",
   "model_type": "chat model, one letter",
   "name": "Claude Fable 5.1 (high)",
   "notes": null,
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 10.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort high, the default (adaptive thinking)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "claude-fable-5-1-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Gemini 4 Argon (off)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Gemini 4 Argon",
   "model_id": "gemini-4-argon-low",
   "model_type": "chat model, one letter",
   "name": "Gemini 4 Argon (low)",
   "notes": "Antigravity model id gemini-4-argon-low (model version gemini-4-argon); list price $2/$10 per million tokens until 2026-12-01 ($4/$20 after)",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 10.0
   },
   "provider": "Google Antigravity subscription",
   "reasoning": "thinking level low, the lowest",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "gemini-4-argon-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Gemini 4 Argon (on)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Gemini 4 Argon",
   "model_id": "gemini-4-argon-high",
   "model_type": "chat model, one letter",
   "name": "Gemini 4 Argon (high)",
   "notes": "Antigravity model id gemini-4-argon-high (model version gemini-4-argon); list price $2/$10 per million tokens until 2026-12-01 ($4/$20 after)",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 10.0
   },
   "provider": "Google Antigravity subscription",
   "reasoning": "thinking level high, Gemini's default",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "temperature 1.0",
   "setting": "high",
   "slug": "gemini-4-argon-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Grok 4.7 (off)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Grok 4.7",
   "model_id": "grok-4.7",
   "model_type": "chat model, one letter",
   "name": "Grok 4.7 (low)",
   "notes": "letter constrained by a JSON schema",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 6.0
   },
   "provider": "xAI subscription (SuperGrok)",
   "reasoning": "reasoning effort low, the lowest grok-4.7 takes (none is rejected); it still reasons",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "grok-4-7-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "aliases": [
    "Grok 4.7 (on)"
   ],
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "Grok 4.7",
   "model_id": "grok-4.7",
   "model_type": "chat model, one letter",
   "name": "Grok 4.7 (high)",
   "notes": "letter constrained by a JSON schema",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 6.0
   },
   "provider": "xAI subscription (SuperGrok)",
   "reasoning": "reasoning effort high, grok-4.7's default",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "grok-4-7-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "leagues/league3.json",
    "tools/v3/page_v3.py",
    "src/players/llm.rs",
    "turns.provider/quantization/probability_source/group_size (snapshot)",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": null,
   "missing": [],
   "model": "GNUbg weak",
   "model_id": null,
   "model_type": "program",
   "name": "GNUbg weak",
   "notes": "GNU Backgammon 1.08, Beginner level (0-ply, deterministic noise 0.060); fixed anchor in every league",
   "option_groups": null,
   "price": null,
   "provider": "GNU Backgammon 1.08, Beginner level (0-ply, noise 0.060), x86-64 build (see About the anchor)",
   "reasoning": null,
   "rules_text": null,
   "runs_on": "CPU",
   "sampling": null,
   "setting": null,
   "slug": "gnubg-weak",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "tools/v3/page_v3.py"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "weights",
    "option_groups",
    "date_run"
   ],
   "model": "Jev 1.13.0",
   "model_id": "jev-1.13.0",
   "model_type": "decision model",
   "name": "Jev 1.13.0",
   "notes": null,
   "option_groups": null,
   "price": {
    "date": "2026-10-08",
    "input_per_mtok": 0.042,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "TypeSafe System One API",
   "reasoning": "none (decision model)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "v1-jev-1-13-0",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "weights",
    "option_groups",
    "date_run"
   ],
   "model": "GPT-6 Luna (Decisions API)",
   "model_id": "gpt-6-luna",
   "model_type": "decision model",
   "name": "GPT-6 Luna (Decisions API)",
   "notes": null,
   "option_groups": null,
   "price": {
    "date": "2026-10-08",
    "input_per_mtok": 0.1,
    "label": "API price",
    "output_per_mtok": 0.0
   },
   "provider": "OpenAI Decisions API",
   "reasoning": "none (decision model)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "v1-gpt-6-luna",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "price.date",
    "date_run"
   ],
   "model": "Perplexity Decider 27B v1.1 (local GPU)",
   "model_id": "pplx-decider-v1.1-27b",
   "model_type": "decision model",
   "name": "Perplexity Decider 27B v1.1 (local GPU)",
   "notes": null,
   "option_groups": "groups of up to 25 options (8,192-token limit)",
   "price": {
    "date": null,
    "input_per_mtok": 0.02,
    "label": "list-price estimate; runs on our GPUs",
    "output_per_mtok": 0.0
   },
   "provider": "our GPUs (released checkpoint, its own server code)",
   "reasoning": "none (decision model)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "v1-decider-v1-1-27b",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "date_run"
   ],
   "model": "Clef",
   "model_id": "clef",
   "model_type": "decision model",
   "name": "Clef",
   "notes": null,
   "option_groups": "groups of up to 110 options",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs",
   "reasoning": "none (decision model)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "v1-clef",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "date_run"
   ],
   "model": "Open-Jev 27B",
   "model_id": null,
   "model_type": "decision model",
   "name": "Open-Jev 27B",
   "notes": "Open-Jev 27B v1.1: a LoRA adapter and decision head on a 27B base model; 4,096-token limit per scored sequence; its server takes no model field",
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs",
   "reasoning": "none (decision model)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "v1-open-jev-27b",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "sampling",
    "date_run"
   ],
   "model": "Cygnet",
   "model_id": "cygnet",
   "model_type": "decision model",
   "name": "Cygnet",
   "notes": "Google's gemma-4-12B-it, unchanged, behind a server that labels the options with letters and reads each letter's probability",
   "option_groups": "groups of up to 26 options",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs (letter-reading server)",
   "reasoning": "none (letter read directly)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": null,
   "setting": null,
   "slug": "v1-cygnet",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "sampling",
    "date_run"
   ],
   "model": "Gemma 4 31B",
   "model_id": "gemma-4-31B-it",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Gemma 4 31B",
   "notes": "calibration setting fitted for Cygnet's 12B model: changes reported confidence, not the play picked",
   "option_groups": "groups of up to 26 options",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs, FP8 (Cygnet's letter-reading server)",
   "reasoning": "none (letter read directly)",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": null,
   "setting": null,
   "slug": "v1-gemma-4-31b",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "reasoning",
    "option_groups",
    "date_run"
   ],
   "model": "DeepSeek-V4.1-Flash",
   "model_id": "deepseek-v4.1-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "DeepSeek-V4.1-Flash",
   "notes": null,
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs (SGLang) behind Cygnet's letter-reading server",
   "reasoning": null,
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": null,
   "slug": "v1-deepseek-v4-1-flash",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "reasoning",
    "option_groups",
    "date_run"
   ],
   "model": "Qwen3.8 Flash Next",
   "model_id": "qwen3.8-flash-next",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Qwen3.8 Flash Next",
   "notes": "Qwen3.8-Flash-Next-FP8",
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs (vLLM, FP8, 4-way tensor parallel) behind Cygnet's letter-reading server",
   "reasoning": null,
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": null,
   "slug": "v1-qwen3-8-flash-next",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "reasoning",
    "option_groups",
    "date_run"
   ],
   "model": "GLM-4.7-Flash",
   "model_id": "glm-4.7-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "GLM-4.7-Flash",
   "notes": null,
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs (vLLM, BF16) behind Cygnet's letter-reading server",
   "reasoning": null,
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": null,
   "slug": "v1-glm-4-7-flash",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "model_id",
    "provider",
    "weights",
    "runs_on",
    "reasoning",
    "option_groups",
    "sampling",
    "price",
    "date_run"
   ],
   "model": "GLM-5.3-Flash NVFP4",
   "model_id": null,
   "model_type": "chat model, answer-letter probabilities",
   "name": "GLM-5.3-Flash NVFP4",
   "notes": null,
   "option_groups": null,
   "price": null,
   "provider": null,
   "reasoning": null,
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": null,
   "sampling": null,
   "setting": null,
   "slug": "v1-glm-5-3-flash-nvfp4",
   "sources": [
    "replay.db answers"
   ],
   "weights": null
  },
  {
   "confidence_source": "none",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "date_run"
   ],
   "model": "Gemini 3.8 Flash",
   "model_id": "gemini-3.8-flash",
   "model_type": "chat model, one letter",
   "name": "Gemini 3.8 Flash",
   "notes": "no token probabilities from this API; answer constrained to the option letters, stored as choice only",
   "option_groups": null,
   "price": {
    "date": "2026-10-14",
    "input_per_mtok": 0.75,
    "label": "API price",
    "output_per_mtok": 3.75
   },
   "provider": "Gemini API",
   "reasoning": "thinking level low",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": null,
   "slug": "v1-gemini-3-8-flash",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "price.date",
    "date_run"
   ],
   "model": "GPT-6 Astra",
   "model_id": "gpt-6-astra",
   "model_type": "chat model, one letter",
   "name": "GPT-6 Astra",
   "notes": "answer constrained to the option letters by a JSON schema; stored as choice only",
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": 10.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50.0
   },
   "provider": "ChatGPT subscription (Codex backend)",
   "reasoning": "reasoning effort low",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": null,
   "slug": "v1-gpt-6-astra",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "price.date",
    "date_run"
   ],
   "model": "Claude Opus 5.5",
   "model_id": "claude-opus-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Opus 5.5",
   "notes": "answer constrained to the option letters by a JSON schema; stored as choice only",
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": 4.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 20.0
   },
   "provider": "Claude subscription",
   "reasoning": "effort low",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": null,
   "slug": "v1-claude-opus-5-5",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "date_run"
   ],
   "model": "DeepSeek-V4.1-Flash",
   "model_id": "deepseek-v4.1-flash",
   "model_type": "chat model, one letter",
   "name": "DeepSeek-V4.1-Flash (thinking)",
   "notes": "answer constrained by a JSON schema; stored as choice only",
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "own hardware",
    "output_per_mtok": null
   },
   "provider": "our GPUs (SGLang)",
   "reasoning": "thinking on, reasoning effort low",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "our GPUs",
   "sampling": "temperature 1.0",
   "setting": "thinking",
   "slug": "v1-deepseek-v4-1-flash-thinking",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "price.date",
    "date_run"
   ],
   "model": "Gemini 4 Argon",
   "model_id": "gemini-4-argon",
   "model_type": "chat model, one letter",
   "name": "Gemini 4 Argon",
   "notes": "no token probabilities; answer constrained to the option letters, stored as choice only",
   "option_groups": null,
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API price",
    "output_per_mtok": 10.0
   },
   "provider": "Gemini API",
   "reasoning": "thinking level low",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": null,
   "slug": "v1-gemini-4-argon",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "v1-set",
   "date_run": null,
   "missing": [
    "option_groups",
    "price.date",
    "date_run"
   ],
   "model": "Gemini 4 Argon",
   "model_id": "gemini-4-argon",
   "model_type": "chat model, one letter",
   "name": "Gemini 4 Argon (shuffled)",
   "notes": "the same setup with the options in a seeded random order per position, to measure the effect of the v1 play order",
   "option_groups": null,
   "play_order": "shuffled (replay of the v1 set with the plays in a seeded random order)",
   "price": {
    "date": null,
    "input_per_mtok": 2.0,
    "label": "API price",
    "output_per_mtok": 10.0
   },
   "provider": "Gemini API",
   "reasoning": "thinking level low",
   "rules_text": "short (the v1 rules text; the same stored request text for every model)",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": null,
   "slug": "v1-gemini-4-argon-shuffled",
   "sources": [
    "tools/replay.py",
    "replay.db answers",
    "analysis/backgammon/index.html",
    "mailbox/Backgammon benchmark v3 design final.md",
    "players.settings (snapshot of v3.db)"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Gemini 3.8 Flash",
   "model_id": "gemini-3.8-flash-low",
   "model_type": "chat model, one letter",
   "name": "Gemini 3.8 Flash (low) [antigravity]",
   "notes": "Antigravity model id gemini-3.8-flash-low",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.75,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 3.75
   },
   "provider": "Google Antigravity subscription",
   "reasoning": "thinking level low",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "smoke-gemini-3-8-flash-low-antigravity",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Claude Haiku 5.5",
   "model_id": "claude-haiku-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Haiku 5.5 (low)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.1,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 0.5
   },
   "provider": "Claude subscription",
   "reasoning": "effort low",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "smoke-claude-haiku-5-5-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Claude Haiku 5.5",
   "model_id": "claude-haiku-5-5",
   "model_type": "chat model, one letter",
   "name": "Claude Haiku 5.5 (high)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.1,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 0.5
   },
   "provider": "Claude subscription",
   "reasoning": "effort high",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "smoke-claude-haiku-5-5-high",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Claude Fable 5.1",
   "model_id": "claude-fable-5-1",
   "model_type": "chat model, one letter",
   "name": "Claude Fable 5.1 (low)",
   "notes": "also in League 3 at the same effort",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 10,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50
   },
   "provider": "Claude subscription",
   "reasoning": "effort low (adaptive thinking only)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "smoke-claude-fable-5-1-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Claude Fable 5.1",
   "model_id": "claude-fable-5-1",
   "model_type": "chat model, one letter",
   "name": "Claude Fable 5.1 (high)",
   "notes": "also in League 3 at the same effort",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 10,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50
   },
   "provider": "Claude subscription",
   "reasoning": "effort high (adaptive thinking only)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "smoke-claude-fable-5-1-high",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Mercury 3",
   "model_id": "inception/mercury-3",
   "model_type": "chat model, one letter",
   "name": "Mercury 3 (off)",
   "notes": "unconstrained letter",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.1,
    "label": "API price",
    "output_per_mtok": 0.35
   },
   "provider": "OpenRouter, pinned to Inception (its only endpoint; quantization not published)",
   "reasoning": "reasoning off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "smoke-mercury-3-off",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Mercury 3",
   "model_id": "inception/mercury-3",
   "model_type": "chat model, one letter",
   "name": "Mercury 3 (low)",
   "notes": "unconstrained letter; reasoning text never returned",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.1,
    "label": "API price",
    "output_per_mtok": 0.35
   },
   "provider": "OpenRouter, pinned to Inception (its only endpoint; quantization not published)",
   "reasoning": "reasoning effort low",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "smoke-mercury-3-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "MiMo V2.6 Pro",
   "model_id": "xiaomi/mimo-v2.6-pro",
   "model_type": "chat model, one letter",
   "name": "MiMo V2.6 Pro (off)",
   "notes": "open weights (RL and MOPD releases); which one the API serves is not published",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.435,
    "label": "API price",
    "output_per_mtok": 0.87
   },
   "provider": "OpenRouter, pinned to Xiaomi (fp8)",
   "reasoning": "reasoning off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "smoke-mimo-v2-6-pro-off",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "MiMo V2.6 Pro",
   "model_id": "xiaomi/mimo-v2.6-pro",
   "model_type": "chat model, one letter",
   "name": "MiMo V2.6 Pro (high)",
   "notes": "open weights (RL and MOPD releases); which one the API serves is not published",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.435,
    "label": "API price",
    "output_per_mtok": 0.87
   },
   "provider": "OpenRouter, pinned to Xiaomi (fp8)",
   "reasoning": "reasoning effort high",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "high",
   "slug": "smoke-mimo-v2-6-pro-high",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Beam 1.0",
   "model_id": "reflection/beam-1.0-501b-a23b",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Beam 1.0 (low)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.25,
    "label": "API price",
    "output_per_mtok": 1.0
   },
   "provider": "OpenRouter, pinned to Nebius (fp8)",
   "reasoning": "reasoning effort low (reasoning is mandatory)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "smoke-beam-1-0-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Beam 1.0",
   "model_id": "reflection/beam-1.0-501b-a23b",
   "model_type": "chat model, answer-letter probabilities",
   "name": "Beam 1.0 (high)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.25,
    "label": "API price",
    "output_per_mtok": 1.0
   },
   "provider": "OpenRouter, pinned to Nebius (fp8)",
   "reasoning": "reasoning effort high (reasoning is mandatory)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "high",
   "slug": "smoke-beam-1-0-high",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "MiMo V2.6 Flash",
   "model_id": "xiaomi/mimo-v2.6-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "MiMo V2.6 Flash (off)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.13,
    "label": "API price",
    "output_per_mtok": 0.28
   },
   "provider": "OpenRouter, pinned to Makora (fp8)",
   "reasoning": "reasoning off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "smoke-mimo-v2-6-flash-off",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "MiMo V2.6 Flash",
   "model_id": "xiaomi/mimo-v2.6-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "MiMo V2.6 Flash (off, unconstrained)",
   "notes": "unconstrained letter",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.13,
    "label": "API price",
    "output_per_mtok": 0.28
   },
   "provider": "OpenRouter, pinned to Makora (fp8)",
   "reasoning": "reasoning off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "smoke-mimo-v2-6-flash-off-unconstrained",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "MiMo V2.6 Flash",
   "model_id": "xiaomi/mimo-v2.6-flash",
   "model_type": "chat model, answer-letter probabilities",
   "name": "MiMo V2.6 Flash (low)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.13,
    "label": "API price",
    "output_per_mtok": 0.28
   },
   "provider": "OpenRouter, pinned to Makora (fp8)",
   "reasoning": "reasoning effort low",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "smoke-mimo-v2-6-flash-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Kimi K3",
   "model_id": "moonshotai/kimi-k3",
   "model_type": "chat model, one letter",
   "name": "Kimi K3 (low)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.239,
    "label": "API price",
    "output_per_mtok": 11.936
   },
   "provider": "OpenRouter, pinned to Inference.net (fp4)",
   "reasoning": "effort low (an effort is required; no off)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "smoke-kimi-k3-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Hy4 Preview",
   "model_id": "tencent/hy4-preview",
   "model_type": "chat model, one letter",
   "name": "Hy4 Preview (off)",
   "notes": null,
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.834,
    "label": "API price",
    "output_per_mtok": 2.501
   },
   "provider": "OpenRouter, pinned to Tencent (fp8)",
   "reasoning": "reasoning off",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "smoke-hy4-preview-off",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "MiniMax-M3.1-Flash",
   "model_id": "minimax/minimax-m3.1-flash:free",
   "model_type": "chat model, one letter",
   "name": "MiniMax-M3.1-Flash (low)",
   "notes": "not on Hugging Face, treated as closed; free endpoint, priced at MiniMax's list price; unconstrained letter",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 0.15,
    "label": "API price",
    "output_per_mtok": 0.6
   },
   "provider": "OpenRouter, MiniMax (free listing, its only one)",
   "reasoning": "effort low (reasoning is mandatory)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "smoke-minimax-m3-1-flash-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Grok 4.7",
   "model_id": "grok-4.7",
   "model_type": "chat model, one letter",
   "name": "Grok 4.7 (low)",
   "notes": "also in League 3 at the same effort",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 2,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 6
   },
   "provider": "xAI subscription (SuperGrok)",
   "reasoning": "reasoning effort low",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "low",
   "slug": "smoke-grok-4-7-low",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "smoke-part-a",
   "date_run": "2026-10-21",
   "missing": [],
   "model": "Grok 4.7",
   "model_id": "grok-4.7",
   "model_type": "chat model, one letter",
   "name": "Grok 4.7 (high)",
   "notes": "also in League 3 at the same effort",
   "option_groups": "one request (no split)",
   "price": {
    "date": "2026-10-21",
    "input_per_mtok": 2,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 6
   },
   "provider": "xAI subscription (SuperGrok)",
   "reasoning": "reasoning effort high, the default",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "smoke-grok-4-7-high",
   "sources": [
    "smoke/part-a.json",
    "mailbox/Part A smoke tests - results.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": "24–28 Oct 2026",
   "missing": [],
   "model": "DeepSeek-V4.1-Flash 1017",
   "model_id": "deepseek/deepseek-v4.1-flash-1017",
   "model_type": "chat model, answer-letter probabilities",
   "name": "DeepSeek-V4.1-Flash 1017 (off)",
   "notes": "The 17 Oct 2026 snapshot of DeepSeek V4.1 Flash, served by DeepSeek itself through OpenRouter (provider pinned, fallbacks off) for every decision; never on our GPUs. Late entry to League 2 on 24 Oct 2026; its missing pairings were played on 27–28 Oct 2026. No fixed answer format (DeepSeek's endpoint takes no JSON schema): the letter is read from the free-text answer with its probabilities. DeepSeek may train on paid prompts; allowed by Can, 24 Oct 2026.",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-24",
    "input_per_mtok": 0.15,
    "label": "API price",
    "output_per_mtok": 0.6
   },
   "provider": "DeepSeek (first party) via OpenRouter, provider pinned, no fallbacks; quantization not published",
   "reasoning": "thinking off (reasoning disabled)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "deepseek-v4-1-flash-1017-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "turns.provider (every decision: openrouter:deepseek)",
    "mailbox/DeepSeek V4.1 Flash 1017 - log.md",
    "freeze/league2-additions.json",
    "src/players/llm.rs"
   ],
   "weights": "open"
  },
  {
   "aliases": [
    "DeepSeek V4.1 Flash 1017 (on)"
   ],
   "confidence_source": "letter probability",
   "dataset": "league-2",
   "date_run": "24–28 Oct 2026",
   "missing": [],
   "model": "DeepSeek-V4.1-Flash 1017",
   "model_id": "deepseek/deepseek-v4.1-flash-1017",
   "model_type": "chat model, answer-letter probabilities",
   "name": "DeepSeek-V4.1-Flash 1017 (low)",
   "notes": "The 17 Oct 2026 snapshot of DeepSeek V4.1 Flash, served by DeepSeek itself through OpenRouter (provider pinned, fallbacks off) for every decision; never on our GPUs. Late entry to League 2 on 24 Oct 2026; its missing pairings were played on 27–28 Oct 2026. No fixed answer format (DeepSeek's endpoint takes no JSON schema): the letter is read from the free-text answer with its probabilities. DeepSeek may train on paid prompts; allowed by Can, 24 Oct 2026.",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-24",
    "input_per_mtok": 0.15,
    "label": "API price",
    "output_per_mtok": 0.6
   },
   "provider": "DeepSeek (first party) via OpenRouter, provider pinned, no fallbacks; quantization not published",
   "reasoning": "thinking on, reasoning effort low",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "low",
   "slug": "deepseek-v4-1-flash-1017-on",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "turns.provider (every decision: openrouter:deepseek)",
    "mailbox/DeepSeek V4.1 Flash 1017 - log.md",
    "freeze/league2-additions.json",
    "src/players/llm.rs"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "league-3",
   "date_run": "24–27 Oct 2026",
   "missing": [
    "weights (not verified for K3.1 Pro)"
   ],
   "model": "Kimi K3.1 Pro",
   "model_id": "moonshotai/kimi-k3.1-pro",
   "model_type": "chat model, one letter",
   "name": "Kimi K3.1 Pro (off)",
   "notes": "Moonshot's Kimi K3.1 Pro through OpenRouter, pinned to Moonshot AI's standard endpoint (mxfp4) for every one of its 1,462 league decisions; never on our GPUs. Late entry to League 3 on 24 Oct 2026 (its two missing pairings were played on 27 Oct). Every request sent reasoning_effort \"none\"; all 1,933 responses reported 0 reasoning tokens and carried no reasoning text. The endpoint returned no token probabilities, so it answers with one letter. Prompts are billed at list price; a separate $30 budget cap applied to Kimi.",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-10-24",
    "input_per_mtok": 2.0,
    "label": "API price",
    "output_per_mtok": 10.0
   },
   "provider": "Moonshot AI's standard endpoint via OpenRouter (endpoint tag moonshotai/mxfp4, quantization mxfp4; not the highspeed endpoint), provider pinned, no fallbacks",
   "reasoning": "reasoning off: every request sent reasoning_effort \"none\" (the lowest effort OpenRouter lists for this model on 2026-10-24); every response reported 0 reasoning tokens",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "off",
   "slug": "kimi-k3-1-pro-off",
   "sources": [
    "players.settings (snapshot of v3.db)",
    "turns.provider/quantization (every decision: openrouter:moonshotai/mxfp4, mxfp4)",
    "exchanges (request reasoning_effort, response usage)",
    "mailbox/Reviewed brief 2026-10-24 - status.md"
   ],
   "weights": null
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "2 Nov 2026",
   "missing": [],
   "model": "SubQ Snap 1.1",
   "model_id": "subq-snap-1.1 (SubQ's default decision model, public preview; it replaced SubQ Snap Preview)",
   "model_type": "decision model",
   "name": "SubQ Snap 1.1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 2 Nov 2026. Probabilities arrive as 32-bit floats and are stored as returned (they sum to 1 only within 32-bit rounding).",
   "option_groups": "up to 255 options per request (as SubQ Snap Preview); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-11-01",
    "input_per_mtok": 0.04,
    "label": "API price (billed); output free",
    "output_per_mtok": 0.0,
    "source": "docs.subq.ai/using-the-api/models"
   },
   "provider": "SubQ Decisions API (api.subq.ai/v1/systemone), direct",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "subq-snap-1-1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries 3 - plan 2026-11-02.md",
    "mailbox/Late entries 3 - status 2026-11-02.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "2 Nov 2026",
   "missing": [],
   "model": "Elephant-1",
   "model_id": "baselabs/elephant-1 @ b3f8a93462fb1051a9de031e94a7ff380e254072 (Hugging Face; post-trained from Qwen3.5-9B)",
   "model_type": "decision model",
   "name": "Elephant-1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 2 Nov 2026. Probabilities come rounded to 6 decimals by the loader and are stored as returned.",
   "option_groups": "up to 120 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16, through its model card's loader (qev's DecisionEngine from OneJev 81ce62f with the checkpoint's serve.patch) behind our HTTP server, one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "base-labs-elephant-1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries 3 - plan 2026-11-02.md",
    "mailbox/Late entries 3 - status 2026-11-02.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "2 Nov 2026",
   "missing": [],
   "model": "Mammoth-1",
   "model_id": "baselabs/mammoth-1 @ 61cec44c281c693992e3d6ebeb44dd3402dfa186 (Hugging Face; post-trained from Qwen3.5-27B)",
   "model_type": "decision model",
   "name": "Mammoth-1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 2 Nov 2026. Probabilities come rounded to 6 decimals by the loader and are stored as returned.",
   "option_groups": "up to 120 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16, through its model card's loader (qev's DecisionEngine from OneJev 81ce62f with the checkpoint's serve.patch) behind our HTTP server, one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "base-labs-mammoth-1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries 3 - plan 2026-11-02.md",
    "mailbox/Late entries 3 - status 2026-11-02.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "6 Nov 2026",
   "missing": [],
   "model": "Drex Max 1.1",
   "model_id": "drex-max-v1.1 (Nace.AI, released 28 Oct 2026; no open weights)",
   "model_type": "decision model",
   "name": "Drex Max 1.1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 6 Nov 2026. Probabilities are stored as returned. Its pair with GPT-6 Luna was not played: GPT-6 Luna's API was out of credit; the pair is outside the requested scope (Ümit, 12 Nov).",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "billed_input_per_mtok": 0.03,
    "billed_label": "billed at Nace's launch price",
    "date": "2026-11-05",
    "input_per_mtok": 0.06,
    "label": "list price; output not billed",
    "output_per_mtok": 0.0,
    "source": "nace.ai/drex and console.nace.ai/docs/drex/reference/pricing"
   },
   "provider": "Nace.AI's hosted System One endpoint (console.nace.ai/v1/systemone), direct",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "drex-max-1-1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries Drex - plan 2026-11-05.md",
    "mailbox/Late entries Drex - status 2026-11-06.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "6 Nov 2026",
   "missing": [],
   "model": "Drex 2.0",
   "model_id": "drex-v2.0 (Nace.AI, pinned id, released 6 Oct 2026)",
   "model_type": "decision model",
   "name": "Drex 2.0",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 6 Nov 2026. Probabilities come rounded to 4 decimals by Nace's API and are stored as returned. Its pair with GPT-6 Luna was not played: GPT-6 Luna's API was out of credit; the pair is outside the requested scope (Ümit, 12 Nov).",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-11-05",
    "input_per_mtok": 0.06,
    "label": "list price (billed); output not billed",
    "output_per_mtok": 0.0,
    "source": "console.nace.ai/docs/drex/reference/pricing"
   },
   "provider": "Nace.AI's hosted System One endpoint (console.nace.ai/v1/systemone), direct",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "drex-2-0",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries Drex - plan 2026-11-05.md",
    "mailbox/Late entries Drex - status 2026-11-06.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "6 Nov 2026",
   "missing": [],
   "model": "Drex Max v1",
   "model_id": "nace-ai/drex-max-v1 @ cac65e1982a0237e996419ba559004522d9a33ac (Hugging Face; Qwen3.8-27B backbone)",
   "model_type": "decision model",
   "name": "Drex Max v1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 6 Nov 2026. Probabilities are stored as returned. Its pair with GPT-6 Luna was not played: GPT-6 Luna's API was out of credit; the pair is outside the requested scope (Ümit, 12 Nov).",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16, through Nace's own drex_max loader (Nace's public GitHub repository, commit e206b27) behind our HTTP server, one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "drex-max-v1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries Drex - plan 2026-11-05.md",
    "mailbox/Late entries Drex - status 2026-11-06.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "6 Nov 2026",
   "missing": [],
   "model": "Drex 2.0",
   "model_id": "nace-ai/drex-v2.0 @ e646674476744e3826caf62b834054151e2aba51 (Hugging Face; 35B total, about 3B active)",
   "model_type": "decision model",
   "name": "Drex 2.0 (local)",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 6 Nov 2026. Probabilities come rounded to 4 decimals by that server and are stored as returned. Its pair with GPT-6 Luna was not played: GPT-6 Luna's API was out of credit; the pair is outside the requested scope (Ümit, 12 Nov).",
   "option_groups": "up to 25 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), bf16, through the serve.py and kev loader its checkpoint ships, one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "drex-2-0-local",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries Drex - plan 2026-11-05.md",
    "mailbox/Late entries Drex - status 2026-11-06.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "7 Nov 2026",
   "missing": [],
   "model": "REAX 1.3",
   "model_id": "decisionmodels/REAX-1.3 @ fa4bf7ee3bb56a8bc421dad090631be1c5845c2d (Hugging Face, Apache-2.0; two parts in FP8 on one GPU)",
   "model_type": "decision model",
   "name": "REAX 1.3",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 7 Nov 2026. Each option's probability is the 50/50 mean of the two parts' probabilities, stored as returned; in the smoke test before the run every probability sum was 1.000000. Its pair with GPT-6 Luna was not played: GPT-6 Luna's API was out of credit; the pair is outside the requested scope (Ümit, 12 Nov). Run note: the server's container ran without Docker's no-new-privileges setting, approved for this run only (the host's Docker refused every container with it); every other protection was kept: all capabilities dropped, read-only root filesystem, non-root user, weights mounted read-only, no outside network.",
   "option_groups": "up to 255 options per request (the server's limit, from the model card); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "NVIDIA RTX PRO 6000 Blackwell (96 GB), in Docker, through reax-engine (github.com/reaxlabs/reax-engine @ ff8f595f149abdd50ea914455eadc74cf97a277f, reax13 mode, its serve.py), one request at a time, no outside network",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "reax-1-3",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries REAX - status 2026-11-07.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "10–11 Nov 2026",
   "missing": [],
   "model": "KnowLine 31B Gen1.5",
   "model_id": "PelaAI/KnowLine-31B-Gen1.5 @ 8e56aa1b236a6dd1064dcfd22ba1e5e360e4c140 (Hugging Face, Apache-2.0; a LoRA fine-tune of google/gemma-4-31B-it merged into bf16 weights)",
   "model_type": "decision model",
   "name": "KnowLine 31B Gen1.5",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 10–11 Nov 2026. It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Each option's probability is a softmax over the option labels, stored as returned. Its first run played 4 matches at once (#933–936) on a server meant for one request at a time; those matches were voided (rows kept, left out of every number) and the pairs replayed from the start, one match at a time.",
   "option_groups": "up to 255 options per request (the front end's limit); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-11-10",
    "input_per_mtok": null,
    "label": "no price (our GPU)",
    "output_per_mtok": null,
    "source": "no list price published for KnowLine-31B-Gen1.5"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10), bf16, through the repo's own serve_knowline.sh (SGLang 0.5.21, then its knowline_server.py front end), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "knowline-31b-gen1-5",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries KnowLine - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "11 Nov 2026",
   "missing": [],
   "model": "Gemini Decision 1",
   "model_id": "google/gemini-decision-1 (OpenRouter; replies name the snapshot google/gemini-decision-1-20261111; closed weights)",
   "model_type": "decision model",
   "name": "Gemini Decision 1",
   "notes": "Google's Gemini Decision 1 through OpenRouter, pinned to Google AI Studio (fallbacks off) for every decision; never on our GPUs. Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 11 Nov 2026. It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1016 aborted; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Two runner builds: dab6da5 for its first 9 matches (#993–#1013, before 21:46:31 UTC), with no probability-map check, and 38024d6 (map-check v2) from 21:46:31 UTC for its last 8 (#1053–#1063). Under map-check v2 a returned probability map is used only if it covers exactly the options shown, every value is finite and in [0, 1], and the values sum to 1 within 1e-6; this sum rule is new (no frozen sum tolerance existed), was chosen on 11 Nov 2026 and applies to this entry only; a failed map still plays the returned choice and stores no probabilities. An offline audit of the first matches under the same rule found no failing map. Option order: we send the options in our seeded order, as for every setup, while Google's docs say the API sorts the option keys alphabetically (UTF-8 byte order; an exact tie goes to the alphabetically first key), so the order the model sees is probably not ours (Google's docs; not verified by us). Its cost is the billed cost of each decision; the entry, with its paid opponents, billed $1.1819 in all.",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-11-11",
    "input_per_mtok": 0.1,
    "label": "API price (billed)",
    "output_per_mtok": 0.0,
    "source": "aistudio.google.com/docs/pricing#gemini-decision and openrouter.ai/google/gemini-decision-1 ($0.10/M input, no output tokens), read 2026-11-11; cost is the billed cost per decision (turns.cost_usd)"
   },
   "provider": "OpenRouter → Google AI Studio: provider pinned to Google AI Studio (only google-ai-studio), no fallbacks; one match at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "gemini-decision-1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/probability_source/cost_usd/harness_commit (v3.db, read-only)",
    "mailbox/Late entry Gemini Decision 1 - status.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "none",
   "dataset": "late-entry",
   "date_run": "10–11 Nov 2026",
   "missing": [],
   "model": "Muse Ember",
   "model_id": "meta/muse-ember (OpenRouter; Meta's endpoint meta/muse-ember-20261109 on its endpoints list of 2026-11-10; closed weights)",
   "model_type": "chat model, one letter",
   "name": "Muse Ember (max)",
   "notes": "Meta's Muse Ember through OpenRouter, pinned to Meta (fallbacks off) for every decision; never on our GPUs. Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup); its matches were played on 10–11 Nov 2026. It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #932 stopped before any decision, unbilled; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Every request sent reasoning_effort \"max\"; 21.76 M of its 22.09 M output tokens were reasoning. The endpoint returned no token probabilities, so it answers with one letter, read from the reply; 46 of its 1,713 answers named no option and a random legal play was made. A quality check (12 Nov) found that 1 of those 46 was an OpenRouter provider error (no answer) scored as invalid, in match #926; the other 45 were real length cuts. A full Run 2 of match #926 (same seed, fixed runner 175141f, its own database) played 6 games to 0 against Random with 0 provider errors; the published numbers keep the original match. Each decision is billed at Meta's list price from the tokens OpenRouter reported.",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": "2026-11-10",
    "input_per_mtok": 0.1,
    "label": "API price (billed)",
    "output_per_mtok": 0.5,
    "source": "dev.meta.ai/models/muse-ember and OpenRouter's endpoints list for meta/muse-ember (Meta: $0.10/M prompt, $0.50/M completion), read 2026-11-10"
   },
   "provider": "OpenRouter (Meta): provider pinned to Meta (only Meta), no fallbacks; every reply came from Meta",
   "reasoning": "reasoning effort max (the top of OpenRouter's listed efforts for this model on 2026-11-10)",
   "rules_text": "long",
   "runs_on": "API",
   "sampling": "temperature 1.0",
   "setting": "max",
   "slug": "muse-ember-max",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/probability_source/cost_usd/reasoning_tokens (v3.db, read-only)",
    "mailbox/Late entries Muse Ember - status 2026-11-10.md"
   ],
   "weights": "closed"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "12, 13 and 14 Nov 2026",
   "missing": [],
   "model": "GLiDE 2B",
   "model_id": "fastino/GLiDE-2B @ 00f122c120cdeed82284106cb77d9bab5a50ad16 (Hugging Face, Fastino Community License)",
   "model_type": "decision model",
   "name": "GLiDE 2B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #41. fastino/GLiDE-2B @ 00f122c on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the server its repository ships (python -m glide.server, GLIDE_MAX_BATCH=1), one request at a time; never on a hosted API. First stage 12 Nov 2026: its 9 hosted and program opponents (#1191–#1199, runner e4ce0e9); local stage 13 Nov, 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1298–#1305, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1200 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). It never truncates: a prompt over GLIDE_MAX_LEN 40,000 tokens is refused (422); League 1's longest stored request (251 options) is 28,323 tokens; nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the server its repository ships (python -m glide.server, GLIDE_MAX_BATCH=1), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "glide-2b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - glide-2b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "12 and 14 Nov 2026",
   "missing": [],
   "model": "GLiDE 0.8B",
   "model_id": "fastino/GLiDE-0.8B @ 9eacdeec79fa48ec52a8d6f54b331376b4bbfaf0 (Hugging Face, Fastino Community License)",
   "model_type": "decision model",
   "name": "GLiDE 0.8B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #42. fastino/GLiDE-0.8B @ 9eacdee on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the server its repository ships (python -m glide.server, GLIDE_MAX_BATCH=1), one request at a time; never on a hosted API. First stage 12 Nov 2026: its 9 hosted and program opponents (#1201–#1209, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1315–#1322, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1210 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). It never truncates: a prompt over GLIDE_MAX_LEN 40,000 tokens is refused (422); League 1's longest stored request (251 options) is 28,323 tokens; nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the server its repository ships (python -m glide.server, GLIDE_MAX_BATCH=1), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "glide-0-8b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - glide-0.8b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "12 and 14 Nov 2026",
   "missing": [],
   "model": "GLiDE 0.4B",
   "model_id": "fastino/GLiDE-0.4B @ 6bda7ab91aeb36dcad6d27828a8871e548c7fc71 (Hugging Face, Fastino Community License)",
   "model_type": "decision model",
   "name": "GLiDE 0.4B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #43. fastino/GLiDE-0.4B @ 6bda7ab on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the server its repository ships (python -m glide.server, GLIDE_MAX_BATCH=1, GLIDE_MAX_LEN=8006), one request at a time; never on a hosted API. First stage 12 Nov 2026: its 9 hosted and program opponents (#1211–#1219, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1323–#1330, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1220 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). An encoder (ModernBERT) whose model would shorten a prompt over 7,999 tokens instead of refusing it, so it gets at most 74 options per request: every League 1 stored request of at most 74 options fits (at most 7,833 tokens with the card's prompt), and longer play lists are split into equal groups of at most 74 and a final among the group winners (the league's split rule, as Winnow and Clef; 678 of League 1's 36,995 decisions, 1.8%, have more than 74 options). The server runs with GLIDE_MAX_LEN=8006, which refuses (422) any prompt the model would shorten, as a backstop. Nothing shortened or refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 74 options per request (its encoder's 7,999-token window); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the server its repository ships (python -m glide.server, GLIDE_MAX_BATCH=1, GLIDE_MAX_LEN=8006), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "glide-0-4b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - glide-0.4b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "11 and 14 Nov 2026",
   "missing": [],
   "model": "KnowLine-0.8B-Gen1",
   "model_id": "PelaAI/KnowLine-0.8B-Gen1 @ 6b27831f7b783c03278bd946d4d8f4f7a98a3798 (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "KnowLine-0.8B-Gen1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #25. PelaAI/KnowLine-0.8B-Gen1 @ 6b27831 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the repo's own serve_knowline.sh (SGLang 0.5.21, then its knowline_server.py front end), one request at a time; never on a hosted API. First stage 11 Nov 2026: its 9 hosted and program opponents (#983–#991, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1331–#1338, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #992 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). It never truncates (SGLang without --allow-auto-truncate refuses over its 262,144-token context); nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the repo's own serve_knowline.sh (SGLang 0.5.21, then its knowline_server.py front end), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "knowline-0-8b-gen1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - knowline-0.8b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "12 and 14 Nov 2026",
   "missing": [],
   "model": "KnowLine-2B-Gen1",
   "model_id": "PelaAI/KnowLine-2B-Gen1 @ 53245d470a4ce316e32fd0d18c24e142c697f40b (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "KnowLine-2B-Gen1",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #40. PelaAI/KnowLine-2B-Gen1 @ 53245d4 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the repo's own serve_knowline.sh (SGLang 0.5.21, then its knowline_server.py front end), one request at a time; never on a hosted API. First stage 12 Nov 2026: its 9 hosted and program opponents (#1181–#1189, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1355–#1362, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1190 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). It never truncates (SGLang without --allow-auto-truncate refuses over its 262,144-token context); nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the repo's own serve_knowline.sh (SGLang 0.5.21, then its knowline_server.py front end), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "knowline-2b-gen1",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - knowline-2b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "11 and 14 Nov 2026",
   "missing": [],
   "model": "KnowLine 4B Gen5.5",
   "model_id": "PelaAI/KnowLine-4B-Gen5.5 @ 779944d2cc8ffa19cef386aee83c83bcb967ade2 (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "KnowLine 4B Gen5.5",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #22. PelaAI/KnowLine-4B-Gen5.5 @ 779944d on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), FP8 at load, through the repo's own serve_knowline.sh (SGLang 0.5.21 with FP8 at load, as the card serves it, then its knowline_server.py front end), one request at a time; never on a hosted API. First stage 11 Nov 2026: its 9 hosted and program opponents (#957–#965, runner b727222); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1371–#1378, runner b727222). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #966 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). FP8 at load (--quantization fp8, as the card and serve_knowline.sh serve it), --mem-fraction-static 0.72. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), FP8 at load, through the repo's own serve_knowline.sh (SGLang 0.5.21 with FP8 at load, as the card serves it, then its knowline_server.py front end), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "knowline-4b-gen5-5",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries KnowLine 4B - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "11 and 14 Nov 2026",
   "missing": [],
   "model": "StartLux-Decision-v1.1-0.8B-Instant",
   "model_id": "startlux-models/StartLux-Decision-v1.1-0.8B-Instant @ 6496ae641cc71b149ce9dad69bd97f2a44e7bf11 (Hugging Face, CC BY-NC 4.0, run for this non-commercial benchmark as the 2B)",
   "model_type": "decision model",
   "name": "StartLux-Decision-v1.1-0.8B-Instant",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #27. startlux-models/StartLux-Decision-v1.1-0.8B-Instant @ 6496ae6 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through its package's own server (startlux-decide 1.1.0 serve --launch-vllm, vLLM 0.31.0), one request at a time; never on a hosted API. First stage 11 Nov 2026: its 9 hosted and program opponents (#1000–#1014, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1339–#1346, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1015 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Instant: one read-out per pass, no thinking; 26 letters per pass including the hidden 'None of the above' (serve's default none_check); the server knocks longer lists out in groups of at most 25. Never truncates: vLLM refuses over 131,072 tokens (422); nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through its package's own server (startlux-decide 1.1.0 serve --launch-vllm, vLLM 0.31.0), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "startlux-decision-v1-1-0-8b-instant",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - startlux-0.8b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "11 and 14 Nov 2026",
   "missing": [],
   "model": "StartLux-Decision-v1.1-2B-Instant",
   "model_id": "startlux-models/StartLux-Decision-v1.1-2B-Instant @ f1e6cd147dda82c9ffa3e8e725d5c1f175566fb1 (Hugging Face, CC BY-NC 4.0)",
   "model_type": "decision model",
   "name": "StartLux-Decision-v1.1-2B-Instant",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #23. startlux-models/StartLux-Decision-v1.1-2B-Instant @ f1e6cd1 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through its package's own server (startlux-decide 1.1.0 serve --launch-vllm, vLLM 0.31.0), one request at a time; never on a hosted API. First stage 11 Nov 2026: its 9 hosted and program opponents (#970–#978, runner 2a27d5d); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1379–#1386, runner 2a27d5d). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #979 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Instant: one read-out per pass, no thinking; one letter per option, 26 per pass; the server knocks longer lists out in groups of at most 25 (the 3 best of each to a final pass); serve's default none_check (a hidden 'None of the above' option). Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through its package's own server (startlux-decide 1.1.0 serve --launch-vllm, vLLM 0.31.0), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "startlux-decision-v1-1-2b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries StartLux 2B - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "13 and 14 Nov 2026",
   "missing": [],
   "model": "StartLux-Decision-v1.1-4B-Instant",
   "model_id": "startlux-models/StartLux-Decision-v1.1-4B-Instant @ 416088f77cfce09633906a3675c7cd2e5dabcf10 (Hugging Face, CC BY-NC 4.0, run for this non-commercial benchmark as the 0.8B and 2B)",
   "model_type": "decision model",
   "name": "StartLux-Decision-v1.1-4B-Instant",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #47. startlux-models/StartLux-Decision-v1.1-4B-Instant @ 416088f on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through its package's own server (startlux-decide 1.1.0 serve --launch-vllm, vLLM 0.31.0), one request at a time; never on a hosted API. First stage 13 Nov 2026: its 9 hosted and program opponents (#1251–#1259, runner 62ae46a); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1387–#1394, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1260 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Instant: one read-out per pass, no thinking; 26 letters per pass including the hidden 'None of the above' (serve's default none_check); the server knocks longer lists out in groups of at most 25. Never truncates: vLLM refuses over 131,072 tokens (422); nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through its package's own server (startlux-decide 1.1.0 serve --launch-vllm, vLLM 0.31.0), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "startlux-decision-v1-1-4b-instant",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - startlux-4b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "11 and 14 Nov 2026",
   "missing": [],
   "model": "OneJev1.6-0.8B",
   "model_id": "OmniJev/OneJev1.6-0.8B @ bcd88a382f314b26cb7b1b7f369c1ed3502ee7a0 (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "OneJev1.6-0.8B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #30. OmniJev/OneJev1.6-0.8B @ bcd88a3 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the server its card names (Flide 0.1.11 flide serve, format onejev-slot-v2), one request at a time; never on a hosted API. First stage 11 Nov 2026: its 9 hosted and program opponents (#1041–#1049, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1347–#1354, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1050 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). No calibration file shipped, so temperature 1; probabilities rounded to 6 places by the server. Never truncates: state plus a question over 32,768 tokens, or a request over 65,536, is refused (422); nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the server its card names (Flide 0.1.11 flide serve, format onejev-slot-v2), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "onejev1-6-0-8b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - onejev1.6-0.8b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "12 and 14 Nov 2026",
   "missing": [],
   "model": "GLiClass v4 Edge",
   "model_id": "knowledgator/gliclass-v4-edge @ 885af1af8a07edf4ed34df83ca0d995ed8201283 (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "GLiClass v4 Edge",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #36. knowledgator/gliclass-v4-edge @ 885af1a on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through turkce's gliclass_server.py with the card's own System One mapping (gliclass 0.1.22, one pass per choice question), one request at a time; never on a hosted API. First stage 12 Nov 2026: its 9 hosted and program opponents (#1141–#1169, runner e4ce0e9); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1363–#1370, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1170 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Options, instructions and state share one 16,384-token input (refused with 422 above it, never cut); League 1's worst case, the longest state plus 218 options at the longest label, is 16,367 tokens, so play lists over 218 options are split into equal groups of at most 218 and a final among the group winners (the league's split rule, as Winnow and Clef; 24 of League 1's 36,995 decisions have more than 218 options). Nothing shortened or refused. Its first stage was interrupted once (12 Nov 08:59 UTC, the server failed 3 health checks in a row) and resumed at 10:15 UTC. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 218 options per request (its 16,384-token input); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through turkce's gliclass_server.py with the card's own System One mapping (gliclass 0.1.22, one pass per choice question), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "gliclass-v4-edge",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - gliclass-v4-edge - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "13 and 14 Nov 2026",
   "missing": [],
   "model": "Cerno 9B v1.2",
   "model_id": "apertixai/cerno-9b-v1.2 @ f02a632fbe3a4698cd1ff5c0b929ea240e2b8eda (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "Cerno 9B v1.2",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #48. apertixai/cerno-9b-v1.2 @ f02a632 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the card's own in-process transformers engine (CernoEngine) behind logic's thin System One server (cerno_server.py, request/response mapping only), one request at a time; never on a hosted API. First stage 13 Nov 2026: its 9 hosted and program opponents (#1261–#1269, runner 2a22d1f); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1395–#1402, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1270 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Probabilities rounded to 4 places. The decider would cut a state ('Context:' block) over 32,768 tokens (its engine default) without refusing; that cut is never reached in League 1 (longest state block 5,856 tokens over its 31,793 distinct stored requests); the longest prompt (251 options) is 28,349 tokens, within its 262,144 context. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the card's own in-process transformers engine (CernoEngine) behind logic's thin System One server (cerno_server.py, request/response mapping only), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "cerno-9b-v1-2",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - cerno-9b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "13 and 14 Nov 2026",
   "missing": [],
   "model": "Turion Jev 31B",
   "model_id": "turion-ai/turion-jev-31b @ 866c0fc6e040cba23fcc72113a80e5c99d5ebab1 (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "Turion Jev 31B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #46. turion-ai/turion-jev-31b @ 866c0fc on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), FP8 at load, through the server its package ships (python -m turion_jev.serve, vLLM 0.32.0 in process, --quantization fp8 --no-recheck), one request at a time; never on a hosted API. First stage 13 Nov 2026: its 9 hosted and program opponents (#1241–#1249, runner a127626 for #1241, then a72223d); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1403–#1410, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1250 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). BF16 weights quantised to FP8 at load and re-scoring off (--no-recheck), as in the card's run; probabilities rounded to 6 places. Its server was stopped after an OOM kill on spark2 (13 Nov 12:33 UTC, 3 failed health checks); from 14:29 UTC it ran through logic's launcher turion_serve_headroom.py --gpu-mem 0.80 (logic #6191), which sets only vLLM's memory share (the KV cache size) and then runs turion_jev.serve unchanged, pinned by its sha256. Never truncates: a prompt over 65,536 tokens is refused (422); nothing refused. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), FP8 at load, through the server its package ships (python -m turion_jev.serve, vLLM 0.32.0 in process, --quantization fp8 --no-recheck), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "turion-jev-31b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - turion-jev-31b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "12 and 14 Nov 2026",
   "missing": [],
   "model": "Sieve 31B v3",
   "model_id": "sthanika-ai/Sieve-31B-v3 @ 8fcee858f7f20a50bfa3c8f984c22adb3cbb46a2 (Hugging Face, Apache-2.0)",
   "model_type": "decision model",
   "name": "Sieve 31B v3",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #44. sthanika-ai/Sieve-31B-v3 @ 8fcee85 on our DGX Spark (spark2, the whole GB10, handed over by logic one server at a time), BF16, through the server its card names (sieve-serve, sieve-decisions 0.5.1 at github.com/sthanika-ai/Sieve 1e64657, defaults), one request at a time; never on a hosted API. First stage 12 Nov 2026: its 9 hosted and program opponents (#1221–#1229, runner a127626); local stage 14 Nov 2026: the 8 setups we run ourselves, one at a time on an RTX PRO 6000 Blackwell of ours (#1411–#1418, runner 2a22d1f). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1230 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Serving policy one read and 2 option orders averaged; probabilities rounded to 4 places. Its replies carry no input-token count, so the runner records 0 input tokens (per-token figures unknown, not zero). Never truncates (no token limit of its own; a request it cannot answer is refused, 422); nothing refused. Its first stage was stopped once after 12 h with pairs open (13 Nov 08:47 UTC) and finished at 11:29 UTC. Its cost is an estimate from its time on the DGX Spark (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA DGX Spark, GB10, spark2), BF16, through the server its card names (sieve-serve, sieve-decisions 0.5.1 at github.com/sthanika-ai/Sieve 1e64657, defaults), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "sieve-31b-v3",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - sieve-31b-v3 - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "14 Nov 2026",
   "missing": [],
   "model": "Liquid d1-2B",
   "model_id": "LiquidAI/d1-2B @ 10e7671d44e4b37700020b1eed3f594c5f2af0b4 (Hugging Face, LFM Open License v1.0; LFM3-VL-2B base)",
   "model_type": "decision model",
   "name": "Liquid d1-2B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #52. LiquidAI/d1-2B @ 10e7671 on one RTX PRO 6000 Blackwell of ours (GPU 2), BF16, through logic's d1_server.py: the command and environment of logic's decision/servers.sh d1-2b (torch 2.15.0+cu132, transformers 5.22.1), the card's call model.system_one(state, questions) unchanged, model field LiquidAI/d1-2B; one request at a time; never on a hosted API. First stage 14 Nov 2026, 16:12–16:16 UTC: its 9 hosted and program opponents (#1419–#1427, runner 8ebad40); local stage 14 Nov 2026, 18:44–19:03 UTC: the 8 setups we run ourselves, one at a time on the same card (#1429–#1436, runner 46c984e, which adds only d1-Omni-2B's entry beside 8ebad40's d1-2B client). It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1428 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Every option count gets distinct single-token codes (A..IU at 255). Context 128,000 positions; nothing in its code shortens a prompt, and a request it cannot answer is refused (400), never cut: League 1's longest stored request (251 options) is 20,759 tokens with its tokenizer and prompt; nothing refused. Its cost is an estimate from its time on the RTX PRO 6000 (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA RTX PRO 6000 Blackwell), BF16, through logic's d1_server.py (the command of logic's decision/servers.sh d1-2b), one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "liquid-d1-2b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - d1-2b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "decision-model probability",
   "dataset": "late-entry",
   "date_run": "14 Nov 2026",
   "missing": [],
   "model": "Liquid d1-Omni-2B",
   "model_id": "LiquidAI/d1-Omni-2B @ 81222dae8bf8bdd7f903643b5b2a8803bf2e3176 (Hugging Face, LFM Open License v1.0; LFM3-Omni-2B base)",
   "model_type": "decision model",
   "name": "Liquid d1-Omni-2B",
   "notes": "Late entry to the original League 1 table (not in standings, not eligible for the Hype Cup), late-entry league #53. LiquidAI/d1-Omni-2B @ 81222da, d1-2B with audio added (the card: text, pictures and audio in), on one RTX PRO 6000 Blackwell of ours (GPU 2, the card d1-2B used, after it), BF16, through logic's d1_server.py: the command of logic's decision/servers.sh d1-omni-2b in d1-2B's environment (torch 2.15.0+cu132, transformers 5.22.1), the card's call model.system_one(state, questions) unchanged, model field LiquidAI/d1-Omni-2B; text only: backgammon positions are text, so its audio and picture inputs are unused; one request at a time; never on a hosted API. Its request and limits are d1-2B's: the model's request code is byte-identical to d1-2B's and its tokenizer only adds two audio tokens, so text prompts tokenize the same. First stage 14 Nov 2026, 19:47–19:50 UTC: its 9 hosted and program opponents (#1437–#1445); local stage 14 Nov 2026, 19:56–20:15 UTC: the 8 setups we run ourselves, one at a time on the same card (#1447–#1454); runner 46c984e throughout. It played 17 of its 19 opponents: its pair with Mercury Decide was not played (Mercury Decide is no longer served on OpenRouter; match #1446 aborted at its warm-up; outside the requested scope, Ümit, 12 Nov) and its pair with GPT-6 Luna was not played (GPT-6 Luna's API was out of credit; outside the requested scope, Ümit, 12 Nov). Every option count gets distinct single-token codes (A..IU at 255). Context 128,000 positions; nothing in its code shortens a prompt, and a request it cannot answer is refused (400), never cut: League 1's longest stored request (251 options) is 20,759 tokens; nothing refused. Its cost is an estimate from its time on the RTX PRO 6000 (GPU time, at most).",
   "option_groups": "up to 255 options per request; more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": null,
    "label": "estimate: GPU time, at most",
    "output_per_mtok": null,
    "source": "content/own_cost.json (wall clock on our machines × hourly rate)"
   },
   "provider": "our GPU (NVIDIA RTX PRO 6000 Blackwell), BF16, through logic's d1_server.py (the command of logic's decision/servers.sh d1-omni-2b), text only, one request at a time",
   "reasoning": "none (decision model)",
   "rules_text": "long",
   "runs_on": "our GPUs",
   "sampling": "not sent (decision request)",
   "setting": null,
   "slug": "liquid-d1-omni-2b",
   "sources": [
    "players.settings (v3.db, read-only)",
    "turns.provider/quantization/probability_source/group_size (v3.db, read-only)",
    "mailbox/Late entries tiny8 - d1-omni-2b - status.md"
   ],
   "weights": "open"
  },
  {
   "confidence_source": "none",
   "dataset": "late-entry",
   "date_run": null,
   "missing": [
    "price.date"
   ],
   "model": "GPT-6 Astra",
   "model_id": "gpt-6-astra",
   "model_type": "chat model, one letter",
   "name": "GPT-6 Astra (high)",
   "notes": "ChatGPT subscription (Codex backend). Late entry to the original League 3 table (not in standings).",
   "option_groups": "up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners",
   "price": {
    "date": null,
    "input_per_mtok": 10.0,
    "label": "API-equivalent (subscription)",
    "output_per_mtok": 50.0
   },
   "provider": "ChatGPT subscription",
   "reasoning": "reasoning effort high (GPT-6 Astra (medium) sent the model's default, medium)",
   "rules_text": "long",
   "runs_on": "subscription",
   "sampling": "not sent (API default)",
   "setting": "high",
   "slug": "gpt-6-astra-high",
   "sources": [
    "players.settings (v3.db, read-only)",
    "the GPT-6 Astra entries of League 3"
   ],
   "weights": "closed"
  }
 ]
}
