GPQA Diamond
Graduate-level science questions written to be Google-proof. Run by Epoch AI.
- Published by
- Epoch AI Benchmarking Hub
- Category
- Reasoning
- Index weight
- 1.0
- Models
- 272
- Data as of
- 9 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
- 1GPT-6 Astra95.8%
- 2Gemini 3.8 Flash95.4%
- 3Gemini 3.7 Flash94.8%
- 4GPT-5.4 Pro94.6%
- 5Gemini 3.1 Pro94.4%
- 6Gemini 3.6 Flash94.1%
- 7Gemini 3.1 Pro94.1%
- 8Grok 4.694.0%
- 9GPT-5.594.0%
- 10GPT-5.5 Pro93.9%
- 11Claude Opus 593.9%
- 12GPT-5.6 Sol93.5%
- 13Grok 4.593.4%
- 14GPT-5.6 Terra93.3%
- 15GPT-5.493.3%
272 of 272
| # | |||||
|---|---|---|---|---|---|
| 1 |  GPT-6 AstraOpenAI | 95.8% | – | 70.6 | – |
| 2 |  Gemini 3.8 FlashGoogle | 95.4% | – | 60.1 | – |
| 3 | Gemini 3.7 FlashGoogle | 94.8% | – | 60.8 | – |
| 4 | GPT-5.4 ProOpenAI | 94.6% | – | 61.1 | – |
| 5 | Gemini 3.1 ProGoogle | 94.4% | – | 60.7 | – |
| 6 | Gemini 3.6 FlashGoogle | 94.1% | – | 58.6 | – |
| 7 | Gemini 3.1 ProGoogle | 94.1% | – | 63.9 | – |
| 8 |  Grok 4.6xAI | 94.0% | – | 58.7 | – |
| 9 | GPT-5.5OpenAI | 94.0% | – | – | – |
| 10 | GPT-5.5 ProOpenAI | 93.9% | – | – | – |
| 11 |  Claude Opus 5Anthropic | 93.9% | – | 66.8 | – |
| 12 | GPT-5.6 SolOpenAI | 93.5% | – | 65.5 | – |
| 13 | Grok 4.5xAI | 93.4% | – | 54.2 | – |
| 14 | GPT-5.6 TerraOpenAI | 93.3% | – | 58.0 | – |
| 15 | GPT-5.4OpenAI | 93.3% | – | 61.3 | – |
| 16 | Grok 4.6xAI | 93.2% | – | 62.3 | – |
| 17 |  Kimi K3Moonshot AIopen | 93.1% | – | 62.7 | – |
| 18 | Claude Opus 5Anthropic | 92.9% | – | 70.0 | – |
| 19 | Gemini 3.5 FlashGoogle | 92.8% | – | 60.4 | – |
| 20 |  Qwen3 8Alibaba | 92.7% | – | 58.0 | – |
| 21 | Gemini 3 ProGoogle | 92.6% | – | 61.6 | – |
| 22 | Kimi K3Moonshot AIopen | 91.9% | – | – | – |
| 23 |  GLM-5.2Zhipu AIopen | 91.9% | – | 57.6 | – |
| 24 |  DeepSeek V4 ProDeepSeekopen | 91.7% | max, 0813 | 56.1 | – |
| 25 | GPT-5.6 LunaOpenAI | 91.6% | – | 56.5 | – |
| 26 | GPT-5.2OpenAI | 91.4% | – | 57.3 | – |
| 27 | Claude Opus 4.8Anthropic | 91.0% | – | 59.4 | – |
| 28 | DeepSeek V4 FlashDeepSeekopen | 91.0% | max, 0731 | 54.7 | – |
| 29 | Deepseek v4 ProDeepSeek | 90.9% | – | 60.3 | – |
| 30 | GLM-5.3Zhipu AIopen | 90.9% | – | 59.9 | – |
| 31 | Qwen3 7Alibaba | 90.9% | – | 58.9 | – |
| 32 |  MiniMax-M3MiniMaxopen | 90.9% | MiniMax-M3 | 56.9 | – |
| 33 | Kimi K2.6Moonshot AIopen | 90.8% | kimi-k2.6 | 60.6 | – |
| 34 | GPT-5.5OpenAI | 90.7% | – | 62.3 | – |
| 35 | Claude Opus 4.6Anthropic | 90.5% | – | 63.1 | – |
| 36 | Claude Sonnet 5Anthropic | 90.5% | – | 57.3 | – |
| 37 | Claude Opus 4.7Anthropic | 90.2% | – | 57.2 | – |
| 38 | GLM-5.3-FlashZhipu AIopen | 90.2% | – | 55.6 | – |
| 39 | GPT-5.6 SolOpenAI | 89.9% | – | 64.6 | – |
| 40 | GLM-5.1Zhipu AIopen | 89.9% | glm-5.1 | 60.1 | – |
| 41 | GPT-5.4OpenAI | 89.9% | – | 58.8 | – |
| 42 |  Muse SparkMeta | 89.8% | – | 65.4 | – |
| 43 | Gemini 3 FlashGoogle | 89.4% | – | 58.5 | – |
| 44 | Grok 4.20xAI | 89.3% | – | 54.3 | – |
| 45 | GPT-5.4OpenAI | 88.9% | – | 56.0 | – |
| 46 | Gemini 3.5 FlashGoogle | 88.9% | – | – | – |
| 47 | Grok 4.3xAI | 88.8% | – | 50.1 | – |
| 48 | Inkling-SmallThinking Machinesopen | 88.5% | – | 48.5 | – |
| 49 | Qwen3.6 PlusAlibabaopen | 88.4% | – | 59.0 | – |
| 50 | Claude Opus 4.6Anthropic | 88.4% | – | 54.1 | – |
| 51 | Claude Opus 4.8Anthropic | 88.4% | – | – | – |
| 52 | GPT-5.2OpenAI | 88.2% | – | 55.5 | – |
| 53 | Claude Opus 5Anthropic | 87.9% | – | 61.7 | – |
| 54 | Qwen3.7 PlusAlibaba | 87.9% | – | 59.7 | – |
| 55 | GPT-5.2OpenAI | 87.9% | – | 58.5 | – |
| 56 | Kimi K2.7 CodeMoonshot AIopen | 87.9% | – | 56.2 | – |
| 57 | GLM-5.2Zhipu AIopen | 87.9% | – | – | – |
| 58 | GLM-5Zhipu AIopen | 87.8% | glm-5 | 59.0 | – |
| 59 | GPT-5.1OpenAI | 87.6% | – | 53.8 | – |
| 60 | Kimi K2.5Moonshot AIopen | 87.6% | fireworks/kimi-k2p5 | 56.1 | – |