Benchmarks · reasoning
benchmarks.bio Expected Performance
Mean IRT-predicted pass rate across automatically discovered evaluations, excluding refusal. Uses the best mapped harness per model and evaluation and requires observed coverage in every domain; the aggregate counts as one benchmark. Score: IRT expected performance across evaluations. Collected by Jake Boggs' benchmark dashboard.
19 models · 19 results · data as of 2026-10-29 · benchmarks.bio ↗
Leaderboard
Opus 5.5 max
All results
19 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Opus 5.5 | max | 55.2% |
| 2 | Sonnet 5.5 | max | 50.7% |
| 3 | Opus 5 | max | 47.4% |
| 4 | GPT-6 Astra | max | 46.8% |
| 5 | Grok 4.6 | max | 42.5% |
| 6 | Grok 4.7 | max | 42.1% |
| 7 | GPT-6 Sol | max | 42.0% |
| 8 | Opus 4.8 | max | 40.7% |
| 9 | GPT-5.6 Sol | max | 39.7% |
| 10 | Gemini 3.7 Flash | max | 37.0% |
| 11 | DeepSeek V4.1 Flash | max | 37.0% |
| 12 | GPT-5.5 | xhigh | 36.4% |
| 13 | Sonnet 5 | max | 36.1% |
| 14 | Gemini 3.8 Flash | max | 35.5% |
| 15 | GPT-6 Luna | max | 32.5% |
| 16 | Kimi K3 | max | 32.4% |
| 17 | Mistral Large 4.0 | high | 29.9% |
| 18 | GPT-5.6 Luna | max | 29.4% |
| 19 | Sonnet 4.6 | max | 28.0% |