Benchmark · Reasoning and knowledge
GPQA Diamond
Multiple-choice questions in biology, chemistry and physics written by PhD-level experts and hard for non-experts to answer even with the web; the Diamond subset is 198 questions. Epoch AI asks zero-shot with chain of thought and scores an answer only in its required format.
- Score
- % of questions correct, higher is better
- Run by
- Epoch AI
- Status
- Current, newest result 19 November 2026
- Updated
- as models are released
- Ceiling
- The best result is within five points of a perfect score, so the test no longer tells the top models apart
- Data
- the published file, CC BY 4.0, read 20 November 2026
Results
45 models we list, each at its best setting, best first.
| # | Model | Score | Range | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|---|
| 1 | GPT-6.2 Astra AA | 96.8% | Independent | effort max | 2026-10-27 | source | |
| 2 | Gemini 4 Hydrogen B | 96.8% | Independent | effort xhigh | 2026-11-13 | source | |
| 3 | Gemini 4 Carbon B | 96.5% | Independent | effort high | 2026-11-02 | source | |
| 4 | Gemini 4 Argon B | 96.4% | Independent | effort high | 2026-10-13 | source | |
| 5 | Gemini 4 Neon B | 96.3% | Independent | effort xhigh | 2026-11-19 | source | |
| 6 | GPT-6 Astra A | 95.8% | Independent | effort max | 2026-08-30 | source | |
| 7 | Claude Sonnet 5.5 A | 95.6% | Independent | effort max | 2026-09-29 | source | |
| 8 | GPT-6.1 Sol AA | 95.4% | Independent | effort max | 2026-09-29 | source | |
| 8 | Gemini 3.8 Flash AA | 95.4% | Independent | effort high | 2026-09-02 | source | |
| 10 | Gemini 3.1 Pro (preview) AA | 94.4% | Independent | effort high | 2026-08-06 | source | |
| 11 | GPT-6.1 Terra AA | 94.3% | Independent | effort max | 2026-10-26 | source | |
| 11 | GPT-6 Sol A | 94.3% | Independent | effort max | 2026-09-22 | source | |
| 13 | Kimi K3.1 Pro D | 94.2% | Independent | effort max | 2026-11-06 | source | |
| 14 | Claude Fable 5.2 A | 93.9% | Independent | effort max | 2026-11-06 | source | |
| 15 | DeepSeek V4.1 Flash C | 93.2% | Independent | effort max | 2026-10-19 | source | |
| 16 | Kimi K3 B | 93.1% | Independent | effort max | 2026-07-16 | source | |
| 17 | Grok 4.7 BB | 92.7% | Independent | effort xhigh | 2026-09-22 | source | |
| 18 | DeepSeek-V4-Flash-0731 C | 91.0% | Independent | effort max | 2026-08-02 | source | |
| 19 | GLM-5.3 B | 90.9% | Independent | effort max | 2026-08-24 | source | |
| 19 | MiniMax M3 C | 90.9% | Independent | 2026-08-10 | source | ||
| 19 | DeepSeek V4 Pro D | 90.9% | Independent | effort high | 2026-08-06 | source | |
| 22 | Claude Opus 5.5 A | 90.6% | Independent | effort max | 2026-09-22 | source | |
| 23 | GPT-6 Luna A | 90.5% | Independent | effort max | 2026-09-22 | source | |
| 24 | Mistral Large 4 C | 90.2% | Independent | effort high | 2026-11-02 | source | |
| 24 | GLM-5.3-Flash B | 90.2% | Independent | effort max | 2026-08-26 | source | |
| 26 | Claude Haiku 5.5 A | 89.7% | Independent | effort max | 2026-10-09 | source | |
| 27 | Grok 4.3 BB | 88.8% | Independent | effort high | 2026-06-17 | source | |
| 28 | Muse Spark Open C | 88.6% | Independent | effort xhigh | 2026-10-15 | source | |
| 29 | Inkling-Small B | 88.5% | Independent | effort xhigh | 2026-08-14 | source | |
| 30 | Inkling B | 88.3% | Independent | effort xhigh | 2026-08-05 | source | |
| 31 | Qwen3.5-397B-A17B B | 86.4% | Independent | effort none | 2026-08-07 | source | |
| 32 | Qwen3.6-27B B | 85.9% | Independent | 2026-08-07 | source | ||
| 33 | Nemotron 3 Ultra 550B A55B B | 85.3% | Independent | 2026-08-10 | source | ||
| 34 | Qwen3.6-35B-A3B B | 84.8% | Independent | effort none | 2026-08-07 | source | |
| 35 | Gemini 3.5 Flash-Lite AA | 83.3% | Independent | effort high | 2026-08-06 | source | |
| 36 | Gemini 3.1 Flash-Lite B | 81.8% | Independent | effort high | 2026-08-06 | source | |
| 37 | Qwen3.5-9B C | 79.0% | Independent | 2026-08-27 | source | ||
| 38 | Gemma 4 31B IT BB | 75.8% | Independent | effort minimal | 2026-08-06 | source | |
| 38 | gpt-oss-120b BB | 75.8% | Independent | effort high | 2025-12-11 | source | |
| 40 | Gemma 4 26B A4B IT BB | 73.2% | Independent | effort minimal | 2026-08-06 | source | |
| 41 | gpt-oss-20b B | 60.8% | Independent | effort medium | 2026-08-27 | source | |
| 42 | GLM-4.7-Flash B | 60.5% | Independent | 2026-08-30 | source | ||
| 43 | Qwen3-30B-A3B-Instruct-2507 C | 55.6% | Independent | 2026-08-30 | source | ||
| 44 | Llama 3.3 70B B | 47.4% | Independent | 2025-01-27 | source | ||
| 45 | Qwen3-4B-Instruct-2507 C | 45.8% | Independent | 2026-08-28 | source |
Every setting, 63 rows
| Model | As the leaderboard names it | Score | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|
| GPT-6.2 Astra | gpt-6.2-astra | 96.8% | Independent | effort max | 2026-10-27 | source |
| Gemini 4 Hydrogen | gemini-4-hydrogen | 96.8% | Independent | effort xhigh | 2026-11-13 | source |
| Gemini 4 Carbon | gemini-4-carbon | 96.5% | Independent | effort high | 2026-11-02 | source |
| Gemini 4 Argon | gemini-4-argon | 96.4% | Independent | effort high | 2026-10-13 | source |
| Gemini 4 Neon | gemini-4-neon | 96.3% | Independent | effort xhigh | 2026-11-19 | source |
| GPT-6 Astra | gpt-6-astra | 95.8% | Independent | effort max | 2026-08-30 | source |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | 95.6% | Independent | effort max | 2026-09-29 | source |
| GPT-6.1 Sol | gpt-6.1-sol | 95.4% | Independent | effort max | 2026-09-29 | source |
| Gemini 3.8 Flash | gemini-3.8-flash | 95.4% | Independent | effort high | 2026-09-02 | source |
| Gemini 3.1 Pro (preview) | gemini-3.1-pro-preview | 94.4% | Independent | effort high | 2026-08-06 | source |
| GPT-6.1 Terra | gpt-6.1-terra | 94.3% | Independent | effort max | 2026-10-26 | source |
| GPT-6 Sol | gpt-6-sol | 94.3% | Independent | effort max | 2026-09-22 | source |
| Kimi K3.1 Pro | kimi-k3.1-pro-together | 94.2% | Independent | effort max | 2026-11-06 | source |
| Gemini 3.1 Pro (preview) | gemini-3.1-pro-preview | 94.1% | Independent | 2026-02-20 | source | |
| Claude Fable 5.2 | claude-fable-5-2 | 93.9% | Independent | effort max | 2026-11-06 | source |
| DeepSeek V4.1 Flash | deepseek-v4.1-flash-1017 | 93.2% | Independent | effort max | 2026-10-19 | source |
| Kimi K3 | kimi-k3 | 93.1% | Independent | effort max | 2026-07-16 | source |
| Grok 4.7 | grok-4.7 | 92.7% | Independent | effort xhigh | 2026-09-22 | source |
| Kimi K3 | kimi-k3 | 91.9% | Independent | effort high | 2026-08-07 | source |
| DeepSeek-V4-Flash-0731 | deepseek-v4-flash-0731 | 91.0% | Independent | effort max | 2026-08-02 | source |
| GLM-5.3 | glm-5.3 | 90.9% | Independent | effort max | 2026-08-24 | source |
| MiniMax M3 | MiniMax-M3 | 90.9% | Independent | 2026-08-10 | source | |
| DeepSeek V4 Pro | deepseek-v4-pro | 90.9% | Independent | effort high | 2026-08-06 | source |
| Claude Opus 5.5 | claude-opus-5-5 | 90.6% | Independent | effort max | 2026-09-22 | source |
| GPT-6 Luna | gpt-6-luna | 90.5% | Independent | effort max | 2026-09-22 | source |
| Mistral Large 4 | mistral-large-4-0-oss | 90.2% | Independent | effort high | 2026-11-02 | source |
| GLM-5.3-Flash | glm-5.3-flash | 90.2% | Independent | effort max | 2026-08-26 | source |
| DeepSeek V4.1 Flash | deepseek-v4.1-flash | 89.8% | Independent | effort max | 2026-10-09 | source |
| Claude Haiku 5.5 | claude-haiku-5-5 | 89.7% | Independent | effort max | 2026-10-09 | source |
| DeepSeek V4 Pro | deepseek-v4-pro | 89.7% | Independent | effort max | 2026-06-16 | source |
| Grok 4.3 | grok-4.3 | 88.8% | Independent | effort high | 2026-06-17 | source |
| Muse Spark Open | Muse-Spark-Open-816B-A42B | 88.6% | Independent | effort xhigh | 2026-10-15 | source |
| Inkling-Small | inkling-small | 88.5% | Independent | effort xhigh | 2026-08-14 | source |
| Inkling | Inkling | 88.3% | Independent | effort xhigh | 2026-08-05 | source |
| Qwen3.5-397B-A17B | qwen3.5-397b-a17b | 86.4% | Independent | effort none | 2026-08-07 | source |
| Qwen3.5-397B-A17B | qwen3.5-397b-a17b | 85.9% | Independent | 2026-08-07 | source | |
| Qwen3.6-27B | qwen3.6-27b | 85.9% | Independent | 2026-08-07 | source | |
| Nemotron 3 Ultra 550B A55B | nemotron-3-ultra | 85.3% | Independent | 2026-08-10 | source | |
| Kimi K3 | kimi-k3 | 84.8% | Independent | effort low | 2026-08-07 | source |
| Qwen3.6-27B | qwen3.6-27b | 84.8% | Independent | effort none | 2026-08-07 | source |
| Qwen3.6-35B-A3B | qwen3.6-35b-a3b | 84.8% | Independent | effort none | 2026-08-07 | source |
| Qwen3.6-35B-A3B | qwen3.6-35b-a3b | 83.8% | Independent | 2026-08-07 | source | |
| Gemini 3.5 Flash-Lite | gemini-3.5-flash-lite | 83.3% | Independent | effort high | 2026-08-06 | source |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite | 81.8% | Independent | effort high | 2026-08-06 | source |
| MiniMax M3 | MiniMax-M3 | 81.3% | Independent | effort none | 2026-08-10 | source |
| Qwen3.5-9B | qwen3.5-9B | 79.0% | Independent | 2026-08-27 | source | |
| Qwen3.5-9B | qwen3.5-9B | 78.9% | Independent | effort none | 2026-08-27 | source |
| Gemini 3.5 Flash-Lite | gemini-3.5-flash-lite | 75.8% | Independent | effort low | 2026-08-06 | source |
| Gemma 4 31B IT | gemma-4-31b-it | 75.8% | Independent | effort minimal | 2026-08-06 | source |
| gpt-oss-120b | openai/gpt-oss-120b | 75.8% | Independent | effort high | 2025-12-11 | source |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite | 74.2% | Independent | effort low | 2026-08-06 | source |
| Gemini 3.5 Flash-Lite | gemini-3.5-flash-lite | 74.2% | Independent | effort minimal | 2026-08-06 | source |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite | 73.7% | Independent | effort minimal | 2026-08-06 | source |
| DeepSeek V4 Pro | deepseek-v4-pro | 73.2% | Independent | effort none | 2026-08-06 | source |
| Gemma 4 26B A4B IT | gemma-4-26b-a4b-it | 73.2% | Independent | effort minimal | 2026-08-06 | source |
| gpt-oss-20b | gpt-oss-20b | 60.8% | Independent | effort medium | 2026-08-27 | source |
| GLM-4.7-Flash | glm-4.7-flash | 60.5% | Independent | 2026-08-30 | source | |
| Qwen3-30B-A3B-Instruct-2507 | qwen3-30b-a3b-instruct-2507 | 55.6% | Independent | 2026-08-30 | source | |
| gpt-oss-20b | gpt-oss-20b | 53.2% | Independent | effort low | 2026-08-27 | source |
| Llama 3.3 70B | Llama-3.3-70B-Instruct | 47.4% | Independent | 2025-01-27 | source | |
| gpt-oss-20b | gpt-oss-20b | 46.0% | Independent | effort high | 2026-08-06 | source |
| Qwen3-4B-Instruct-2507 | qwen3-4b-instruct-2507 | 45.8% | Independent | 2026-08-28 | source | |
| GLM-4.7-Flash | glm-4.7-flash | 45.1% | Independent | effort none | 2026-08-30 | source |
A rank is among the models we list with a result, not a place on the whole leaderboard, and a model offered by several hosts is shown once, on its own listing. These results don't move our grades. Data: Epoch AI, Capabilities and benchmarking (CC BY 4.0), CC BY 4.0.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.