GPQABrowse 284 families
GPQA
Can the model reason through specialist biology, physics, and chemistry questions?
Version / measure
- Domain
- Science
- Ranked models
- 27
- Reported values
- 41
- Reports
- 7 · lab
- Direction
- Higher is better
- Status
- Canonical static set
Rankings
Best reported score per model across effort settings
| # | Model | Best setting | Score |
|---|---|---|---|
| 1 | Claude Mythos PreviewAnthropic | reportedcost n/a1 value · Jul 9 · source | 94.6% |
| 2 | GPT-5.6 SolOpenAI | reportedcost n/a2 values · Jul 9 · source | 94.6% |
| 3 | Gemini 3.1 Pro PreviewGoogle | reportedcost n/a1 value · Jul 9 · source | 94.3% |
| 4 | Gemini 3.1 ProGoogle | reportedcost n/a3 values · Jun 16 · source | 94.3% |
| 5 | Claude Mythos 5Anthropic | reportedcost n/a1 value · Jul 9 · source | 94.1% |
| 6 | GPT-5.5OpenAI | reportedcost n/a3 values · Jul 9 · source | 93.6% |
| 7 | Claude Opus 4.8Anthropic | reportedcost n/a3 values · Jun 16 · source | 93.6% |
| 8 | Kimi K3Moonshot AIOW | maxcost n/a2 values · Jul 17 · source | 93.5% |
| 9 | MiniMax M3MiniMaxOW | reportedcost n/a1 value · Jun 16 · source | 93.0% |
| 10 | GPT-5.6 TerraOpenAI | reportedcost n/a1 value · Jul 9 · source | 92.9% |
Effort curve
Score vs. reported cost. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology
Can the model reason through specialist biology, physics, and chemistry questions?
Multiple-choice science questions written and validated by domain experts. Higher is better. The value is the percentage reported in this lab's table.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
DeepSeek's official V4.1-Flash release table; exact source label GPQA Diamond; harness and effort are not printed. No tools.
Method / source ↗