Benchmarks
GPQA Diamond
Models scored
199
Best score
0.95
Metric
mean_score
GPQA Diamond is an AI evaluation tracked by Epoch AI, with 199 scored model versions on record. Epoch AI runs this evaluation directly, so the scores are reproducible against their published logs.
Scored models span March 2023 to July 2026. The best score recorded is 0.95, measured as mean_score. Models come from Anthropic, OpenAI, Moonshot, DeepSeek, Google, xAI, Z.ai (Zhipu AI), Alibaba and others.
Full record
- Score metric
- mean_score
- Models scored
- 199
- Best score recorded
- 0.95
- Earliest model scored
- Mar 14, 2023
- Most recent model scored
- Jul 24, 2026
- Third-party reported
- No
Leaderboard
| 01 | gpt-5.4-pro-2026-03-05_xhigh | OpenAI | 0.95 |
| 02 | gemini-3.1-pro-preview | Google DeepMind | 0.94 |
| 03 | gpt-5.5-pre-release_xhigh | OpenAI | 0.94 |
| 04 | gpt-5.5-pro-pre-release_xhigh | OpenAI | 0.94 |
| 05 | claude-opus-5_max | Anthropic | 0.94 |
| 06 | gpt-5.6-sol_max | OpenAI | 0.93 |
| 07 | grok-4.5_high | xAI | 0.93 |
| 08 | gpt-5.6-terra_max | OpenAI | 0.93 |
| 09 | gpt-5.4-2026-03-05_xhigh | OpenAI | 0.93 |
| 10 | kimi-k3_max | Moonshot | 0.93 |
| 11 | gemini-3.5-flash_high | 0.93 | |
| 12 | gemini-3-pro-preview | Google DeepMind | 0.93 |
| 13 | glm-5.2_max | Z.ai (Zhipu AI) | 0.92 |
| 14 | qwen3.7-max | Alibaba | 0.92 |
| 15 | gpt-5.6-luna_max | OpenAI | 0.92 |
| 16 | gpt-5.2-2025-12-11_xhigh | OpenAI | 0.91 |
| 17 | claude-opus-4-8_max | Anthropic | 0.91 |
| 18 | kimi-k2.6 | Moonshot | 0.91 |
| 19 | gpt-5.5_low | OpenAI | 0.91 |
| 20 | claude-sonnet-5_xhigh | Anthropic | 0.91 |
| 21 | claude-opus-4-6_32K | Anthropic | 0.91 |
| 22 | claude-opus-4-7_xhigh | Anthropic | 0.9 |
| 23 | gpt-5.4-2026-03-05_high | OpenAI | 0.9 |
| 24 | muse-spark | Meta AI | 0.9 |
| 25 | deepseek-v4-pro_max | DeepSeek | 0.9 |
| 26 | kimi-k2.7-code | Moonshot | 0.9 |
| 27 | grok-4.20-0309-reasoning | xAI | 0.89 |
| 28 | qwen3.6-max-preview | Alibaba | 0.89 |
| 29 | gpt-5.4-2026-03-05_medium | OpenAI | 0.89 |
| 30 | grok-4.3_high | xAI | 0.89 |
| 31 | claude-opus-4-6_64K | Anthropic | 0.89 |
| 32 | gpt-5.2-2025-12-11_high | OpenAI | 0.88 |
| 33 | gpt-5.2-2025-12-11_medium | OpenAI | 0.88 |
| 34 | glm-5 | Z.ai (Zhipu AI) | 0.88 |
| 35 | gpt-5.1-2025-11-13_high | OpenAI | 0.88 |
| 36 | fireworks/kimi-k2p5 | Moonshot | 0.88 |
| 37 | qwen3.6-plus | Alibaba | 0.87 |
| 38 | claude-sonnet-4-6_32K | Anthropic | 0.87 |
| 39 | grok-4-0709 | xAI | 0.87 |
| 40 | gemini-3.5-flash_minimal | 0.86 | |
| 41 | gpt-5-2025-08-07_high | OpenAI | 0.86 |
| 42 | claude-opus-4-5-20251101_32K | Anthropic | 0.86 |
| 43 | glm-5.1 | Z.ai (Zhipu AI) | 0.85 |
| 44 | claude-opus-4-5-20251101_16K | Anthropic | 0.85 |
| 45 | gpt-5-2025-08-07_medium | OpenAI | 0.85 |
| 46 | gemini-2.5-pro | Google DeepMind | 0.85 |
| 47 | gpt-5.1-2025-11-13_medium | OpenAI | 0.85 |
| 48 | gpt-5.4-2026-03-05_low | OpenAI | 0.85 |
| 49 | gemini-2.5-pro-preview-06-05 | Google DeepMind | 0.85 |
| 50 | qwen3.6-flash | Alibaba | 0.84 |
| 51 | kimi-k2-thinking-turbo | Moonshot | 0.84 |
| 52 | qwen3.5-plus | Alibaba | 0.84 |
| 53 | qwen3.5-flash | Alibaba | 0.84 |
| 54 | gemini-2.5-pro-exp-03-25 | Google DeepMind | 0.84 |
| 55 | gpt-5.4-mini-2026-03-17_high | OpenAI | 0.84 |
| 56 | deepseek-reasoner | DeepSeek | 0.83 |
| 57 | claude-sonnet-4-6_medium | Anthropic | 0.83 |
| 58 | claude-sonnet-4-6_high | Anthropic | 0.83 |
| 59 | glm-4.7 | Z.ai (Zhipu AI) | 0.83 |
| 60 | gemini-3-flash-preview | Google DeepMind | 0.83 |
More benchmarks
SourceEpoch AI, 'AI Benchmarking Hub'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/benchmarks. Licensed under CC BY 4.0.
← All benchmarks