Decision model
Perplexity Decider 27B v1.2 API
1 question per request
- Accuracy, all 2,198 questions
- 66.3%64.3% to 68.3%, 95% interval; 1,458 right
- Answers kept at 99% confidence
- 327 of 2,1980 wrong; share right 98.8% to 100.0%, 95% interval
- Cost, $ per 1,000 questionsper 1M questions
- $0.008$7.70Basis and price under Cost
- Latency
- 0.254 smedian request, 1 question per request
The same model with Perplexity's web search on is Perplexity Decider 27B v1.2 · API · with web: Web grounding compares the two.
How we called it
- Kind
- Decision model
- Access
- Perplexity Decisions API
- Paid as
- Provider service
- Model ID
pplx-decider-v1.2- Parameters
- 27B dense, as the maker states it (counted from the released files: 26.09B).language model (no LM head) + 0.46B vision encoder + 255-way readout; source: counted from the checkpoint's safetensors, perplexity-ai/pplx-decider-v1.2@225c913 (the same weights as cce0f3f), huggingface.co/perplexity-ai/pplx-decider-v1.2; read 9 November 2026
- Note
- Perplexity's hosted Decider v1.2 through its Decisions API, web grounding off. The other Decider v1.2 rows are Perplexity's released checkpoint on our hardware and this model with web grounding on.
- Confidence from
- Option score
- Run
- 633, 5 November 2026
Settings
- Reasoning
- off
- Request format
- 1 question per request
- Requests at once
- 1
- Temperature
- not recorded
Run record: harness seed, harness commit and prompt version
- Harness seed
- 0
- Harness commit
b0c7b9874aab- Prompt version
254597b8be16
Results
| Questions | Accuracy | 95% interval | Right |
|---|---|---|---|
| All questions | 66.3% | 64.3% to 68.3% | 1,458 of 2,198 |
| Reading | 89.0% | 86.4% to 91.2% | 584 of 656 |
| Grammar | 56.7% | 54.2% to 59.1% | 874 of 1,542 |
Comparison group: DEFGHIJK. Compare with another setup.
Accuracy by topic
English
The Text tab's 547 kept reading questions translated into English, asked with this setup's Text tab settings. Every setup: English reading; against Turkish: English and Turkish.
- Right answers
- 484 of 547, first English run (637)
- Accuracy
- 88.5%
- Non-hallucination with “I don't know”
- 4.5%: 3 “I don't know” of 66 not right (run 643)
- Answered with “I don't know”
- 99.5%: 544 of 547
- Error rate with “I don't know”
- 11.6%: 63 of 544 answers
- Cost, $ per 1,000 questionsper 1M questions, first English run
- 0.0087.79at list prices
- Latency, median request, first English run
- 0.256 s; 1 request at once
- English run
- 637
- Machine, English runs
- hosted API
- Date
- 5 November 2026
- Request format
- 1 question per request
- Reasoning
- off
- Prompt version
47f5776e71db- Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68
Confidence
- Answers kept at 99% or more
- 327, of which 0 wrong (share right: 98.8% to 100.0%, 95% interval); 1,871 skipped, the model less than 99% sure
- Calibration gap
- 4.2 points
- AUROC, all questions
- 0.84 0.83 to 0.86
- AUROC, reading
- 0.92 0.90 to 0.94
- AUROC, grammar
- 0.78 0.76 to 0.80
- Brier score
- 0.154
- Accuracy of its surest answers
- most confident fifth: 99.8% right; most confident half: 90.5% right; most confident four fifths: 75.0% right
- Typical score of a right answer
- 90.1%
- Typical score of a wrong answer
- 48.0%
Repeats
We did not repeat this setup.
Latency
- Latency, median request
- 0.254 s (this run)
- Latency, 90th percentile
- 0.469 s
- Requests at once
- 1
- Throughput
- 176.5 questions per minute of request time
- Request time
- 12.5 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
- Cost per 1,000 questionsper 1M questions
- $0.008$7.70
- Whole run
- $0.017
- Basis
- Estimated from published prices and recorded token usage
- Price
- $0.02 input, output free per 1M tokens (source, 2026-11-05)
- Input tokens per question
- 385
- Output tokens per question
- 1
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 Flash high · chat | 20.3% | −0.11 points | 20% |
| GPT-6 Astra high · chat | 22.0% | −0.13 points | 22% |
| GPT-6.1 Sol high · chat | 23.0% | −0.16 points | 23% |
| GPT-6 Astra low · chat | 22.8% | −0.16 points | 23% |
| GPT-6 Astra low · chat · 1 per request | 21.0% | −0.14 points | 21% |
| Gemini 3.8 Flash low · chat · 1 per request | 21.3% | −0.15 points | 21% |
| GPT-6.1 Sol low · chat · 1 per request | 22.5% | −0.16 points | 22% |
| GPT-6.1 Sol low · chat | 22.8% | −0.16 points | 22% |
| Claude Opus 5.5 high · chat | 23.5% | −0.20 points | 23% |
| GPT-6.1 Terra high · chat | 23.9% | −0.18 points | 22% |
| Claude Opus 5.5 low · chat · 1 per request | 23.4% | −0.17 points | 23% |
| Gemini 3.8 Flash low · chat | 22.6% | −0.20 points | 20% |
| GPT-6.1 Terra low · chat | 27.3% | −0.19 points | 25% |
| Claude Opus 5.5 low · chat | 30.0% | −0.20 points | 30% |
| Claude Sonnet 5.5 high · chat · 1 per request | 32.1% | −0.14 points | 32% |
| Claude Sonnet 5.5 high · chat | 49.8% | −0.13 points | 50% |