Reasoning
GPQA Diamond
198 graduate-level physics, chemistry and biology questions written to be Google-proof. PhD holders in the matching field score about 65%.
GPQA Diamond is scored as a percentage of questions answered correctly. 83 of the 159 models in this index report a score for it. The highest published figure here is 91.9%, from Gemini 3 Pro.
Reported scores
Top 20 on GPQA Diamond
Ordered by the figure each maker published. Models that have not reported this benchmark are not listed — an absent score is not a low score.
| # | Model | Published score |
|---|---|---|
| 1 | Gemini 3 ProGoogle DeepMind | |
| 2 | Grok 4xAI | |
| 3 | Claude Opus 4.5Anthropic | |
| 4 | GPT-5OpenAI | |
| 5 | Grok 4 FastxAI | |
| 6 | Qwen3-MaxAlibaba Qwen | |
| 7 | Kimi K2 ThinkingMoonshot AI | |
| 8 | Gemini 2.5 ProGoogle DeepMind | |
| 9 | Claude Sonnet 4.5Anthropic | |
| 10 | o3OpenAI | |
| 11 | GLM-4.6Z.ai (Zhipu) | |
| 12 | GPT-5 miniOpenAI | |
| 13 | o4-miniOpenAI | |
| 14 | Claude Opus 4.1Anthropic | |
| 15 | gpt-oss-120bOpenAI | |
| 16 | DeepSeek-V3.1DeepSeek | |
| 17 | DeepSeek-V3.2DeepSeek | |
| 18 | o3-miniOpenAI | |
| 19 | GLM-4.5Z.ai (Zhipu) | |
| 20 | Gemini 2.5 FlashGoogle DeepMind |
Where these numbers come from
Every score on this page is a published figure, taken from the model's own card, system card, technical report or release post, or from a public leaderboard. CorX Labs did not run these evaluations. Most are self-reported by the lab that built the model, which means they were produced under that lab's own choice of prompt, scaffold and number of attempts — so treat them as a starting point for a shortlist, not as a settled ranking.
A score someone other than the model's maker measured is marked Independent and names its measurer. Those are the stronger numbers on this page — an outside harness has no reason to flatter anyone — and there are not many of them.
Where a figure has not been published, the cell reads Not reported rather than an estimate. Nothing here is inferred, interpolated or guessed. Each model records the month its row was last checked. Full method and caveats.
How to read this score
GPQA Diamond is the closest thing to a pure reasoning test in common use. The questions were written by domain PhDs specifically so that a non-expert with unlimited web access still scores around 34%, which means a high score cannot come from retrieval alone.
Where it is weak
There are only 198 questions, so a few lucky answers move the number by a full point. It also covers three sciences and nothing else — it says little about writing, code or judgement.