GPQA Diamond
Epoch AI / Knowledge · Graduate-level physics, chemistry and biology knowledge
Across e/accCounts toward rankingNear saturation
Evidence budget1% (near saturation; original 2%)
Upstream data time10/16 02:52
Last successful sync10/17 08:05
What it measures and how
Use only mean_score from Epoch’s own reruns, retaining reasoning tier and standard error. The 198 science questions require reasoning but do not represent all world knowledge.
How this evidence is used
Counts toward ranking; benchmarks in the same task family share a fixed budget.
Benchmark results
Fixed rules select one representative configuration per public model. Anonymous test models are omitted. Original source ranks are retained, with up to 30 models per benchmark.
| Source rank | Source model name | Raw result | Representative configuration |
|---|---|---|---|
| 1 | gemini-4-argon_highGoogle | 96.4% | High reasoning |
| 2 | gpt-6-astra_maxOpenAI | 95.8% | Max reasoning |
| 3 | claude-sonnet-5-5_maxAnthropic | 95.6% | Max reasoning |
| 4 | gemini-3.8-flash_highGoogle | 95.4% | High reasoning |
| 4 | gpt-6.1-sol_maxOpenAI | 95.4% | Max reasoning |
| 6 | gemini-3.7-flash_highGoogle | 94.8% | High reasoning |
| 7 | gpt-5.4-pro-2026-03-05_xhighOpenAI | 94.6% | xHigh reasoning |
| 8 | gemini-3.1-pro-preview_highGoogle | 94.4% | High reasoning |
| 9 | gpt-6-sol_maxOpenAI | 94.3% | Max reasoning |
| 10 | gemini-3.6-flash_highGoogle | 94.1% | High reasoning |
| 15 | claude-opus-5_maxAnthropic | 93.9% | Max reasoning |
| 16 | gpt-5.6-sol_maxOpenAI | 93.5% | Max reasoning |
| 17 | grok-4.5_highxAI | 93.4% | High reasoning |
| 18 | gpt-5.6-terra_maxOpenAI | 93.3% | Max reasoning |
| 19 | gpt-5.4-2026-03-05_xhighOpenAI | 93.3% | xHigh reasoning |
| 20 | grok-4.6_xhighxAI | 93.2% | xHigh reasoning |
| 21 | kimi-k3_maxMoonshot AI | 93.1% | Max reasoning |
| 23 | gemini-3.5-flash_highGoogle | 92.8% | High reasoning |
| 24 | grok-4.7_xhighxAI | 92.7% | xHigh reasoning |
| 25 | qwen3.8-max_xhighAlibaba | 92.7% | xHigh reasoning |
| 26 | gemini-3-pro-previewGoogle | 92.6% | Source default configuration |
| 27 | qwen3.8-max-0902_xhighAlibaba | 92.3% | xHigh reasoning |
| 29 | glm-5.2_maxZ.ai | 91.9% | Max reasoning |
| 30 | deepseek-v4-pro-0813_maxDeepSeek | 91.7% | Max reasoning |
| 31 | gpt-5.6-luna_maxOpenAI | 91.6% | Max reasoning |
| 32 | gpt-5.2-2025-12-11_xhighOpenAI | 91.4% | xHigh reasoning |
| 33 | claude-opus-4-8_maxAnthropic | 91.0% | Max reasoning |
| 33 | deepseek-v4-flash-0731_maxDeepSeek | 91.0% | Max reasoning |
| 35 | glm-5.3_maxZ.ai | 90.9% | Max reasoning |
| 35 | MiniMax-M3MiniMax | 90.9% | Source default configuration |
Limitations and attribution
Counts toward ranking; benchmarks in the same task family share a fixed budget.
Data license: CC BY 4.0 · Data evaluated by Epoch AI itself
Results are published by Epoch AI. Raw scores and the e/acc consensus index use different scales and cannot be added directly.