Decision model
Perplexity Decider 27B v1.2
local · 1 question per request
67.1% of the 2,198 questions right (65.1% to 69.0%, 95% interval). Back to the leaderboard.
How we called it
- Kind
- Decision model
- Access
- Perplexity Decider server, self-hosted
- Paid as
- Our own hardware
- Model ID
pplx-decider-v1.2- Variant
- local
- Parameters
- 26.1B
- Hardware
- 1 × DGX Spark, GB10 (128 GB unified memory); BF16
- Note
- Perplexity's released checkpoint, run on our hardware. An API run is pending; this is an additional local setup.
- Confidence from
- Option score
- Run
- 217, 17 October 2026
Settings
- Reasoning
- off
- Request format
- 1 question per request
- Requests at once
- 1
- Temperature
- not recorded
- Harness seed
- 0
- Harness commit
f9f71e0fa923- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 67.1% | 65.1% to 69.0% | 1,474 of 2,198 |
| Reading | 89.2% | 86.6% to 91.3% | 585 of 656 |
| Grammar | 57.7% | 55.2% to 60.1% | 889 of 1,542 |
Comparison group: pqr. Compare with another setup.
Accuracy by topic
Page images (preliminary)
Preliminary: these numbers are not final and may change. This setup is condition A of the Page images tab, which asks the 763 questions with underlined or numbered words as text (A), as a page image (B), as both (C) and through OCR models (D); the share answered right, with its 95% interval.
| Model | A · text onlyaccuracy, higher is better | B · image onlyaccuracy, higher is better | C · text plus imageaccuracy, higher is better | Best OCR routeaccuracy, higher is better |
|---|---|---|---|---|
| Perplexity Decider 27B v1.2local · 1 question per request | 62.3%58.8% to 65.6% | No image input | No image input | 58.2%via Surya OCR 254.7% to 61.6% |
Confidence
- Answers kept at 99% or more
- 360, of which 0 wrong (share right: 98.9% to 100.0%, 95% interval); 1,838 skipped, the model less than 99% sure
- Calibration gap
- 4.4 points
- AUROC, all questions
- 0.84 0.82 to 0.85
- AUROC, reading
- 0.92 0.90 to 0.95
- AUROC, grammar
- 0.77 0.75 to 0.80
- Brier score
- 0.156
- Accuracy of its surest answers
- most confident fifth: 99.3% right; most confident half: 90.5% right; most confident four fifths: 75.6% right
- Typical score of a right answer
- 91.2%
- Typical score of a wrong answer
- 49.8%
Repeats
Run 5 times on all questions.
- Answers changed in 5 runs
- 0 of 2,198
- Same wrong answer in all 5
- 724 of 2,198
- Wrong in at least one run
- 724 of 2,198
- Accuracy in each run
- 67.1%, 67.1%, 67.1%, 67.1%, 67.1%
Latency
- Latency, median request
- 0.580 s (this run)
- Latency, 90th percentile
- 0.787 s
- Requests at once
- 1
- Throughput
- 97.4 questions per minute of request time
- Request time
- 22.6 minutes, over the requests whose answers were kept
Latency: the time from sending one request to receiving its answer, measured from our computer, network and any retries within the request included. Request time: the time at least one request was in flight, over the requests whose answers were kept. It is not the run's elapsed time: it leaves out requests that failed and any pause before a run was resumed.
Cost
- $ per 1,000 questions
- $0.008
- Whole run
- $0.017
- Basis
- Estimated at Perplexity's list price for the same model; our hardware cost is not measured.
- Price
- $0.02 input, output free per 1M tokens, plus 3 billed input tokens per request (source, 2026-10-18)
- Input tokens per question
- 385
- Output tokens per question
- 0
Going first in front of an LLM
Replay of recorded answers, cutoff chosen on half the questions, averaged over the splits. Savings are estimates from API prices; the Method page explains how.
| LLM behind it | Share taken | Accuracy change | Saving |
|---|---|---|---|
| Gemini 3.8 Flashhigh reasoning · 16 questions per request | 19.1% | −0.09 points | 19%§ |
| GPT-6 Astrahigh reasoning · 16 questions per request | 19.4% | −0.11 points | 19%§ |
| GPT-6.1 Solhigh reasoning · 16 questions per request | 19.7% | −0.12 points | 19%§ |
| GPT-6 Astralow reasoning · 16 questions per request | 19.5% | −0.11 points | 19%§ |
| GPT-6 Astralow reasoning · 1 question per request | 19.1% | −0.09 points | 19%§ |
| Gemini 3.8 Flashlow reasoning · 1 question per request | 19.2% | −0.10 points | 18%§ |
| GPT-6.1 Sollow reasoning · 1 question per request | 19.6% | −0.12 points | 19%§ |
| GPT-6.1 Sollow reasoning · 16 questions per request | 19.5% | −0.11 points | 19%§ |
| Claude Opus 5.5high reasoning · 16 questions per request | 19.7% | −0.13 points | 20%§ |
| GPT-6.1 Terrahigh reasoning · 16 questions per request | 20.1% | −0.13 points | 19%§ |
| Claude Opus 5.5low reasoning · 1 question per request | 19.7% | −0.13 points | 20%§ |
| Gemini 3.8 Flashlow reasoning · 16 questions per request | 19.7% | −0.12 points | 17%§ |
| GPT-6.1 Terralow reasoning · 16 questions per request | 24.0% | −0.22 points | 22%§ |
| Claude Opus 5.5low reasoning · 16 questions per request | 24.6% | −0.23 points | 24%§ |
| Claude Sonnet 5.5high reasoning · 1 question per request | 27.1% | −0.23 points | 27%§ |
| Claude Sonnet 5.5high reasoning · 16 questions per request | 49.4% | −0.15 points | 49%§ |
§ Counts the first model at its maker’s list price for the same model; our hardware cost is not measured.