Reasoning
GPQA Diamond
Graduate-level multiple-choice questions in biology, chemistry and physics, written by experts to be hard to answer with a web search. Random guessing scores 25%.
Published by Epoch AI under CC BY 4.0. Compare it with other tests
Top 15
Each model at its best configuration, coloured by maker. The dashed line marks what random guessing would score.
The best model from each maker
Each maker’s highest-scoring model on this test, and its rank.
| Maker | Best model | Score | Rank |
|---|---|---|---|
 OpenAI | GPT-6 Astra (max) | 95.8% | #1 of 222 |
 Anthropic | Claude Sonnet 5.5 (max) | 95.6% | #2 of 222 |
 Google | Gemini 3.8 Flash (high) | 95.4% | #3 of 222 |
 xAI | Grok 4.6 (high) | 94.0% | #11 of 222 |
 Moonshot AI | Kimi K3 (max) | 93.1% | #19 of 222 |
 Alibaba | Qwen3.8 Max (xhigh) | 92.7% | #22 of 222 |
 Z.ai | GLM-5.2 (max) | 91.9% | #25 of 222 |
 DeepSeek | DeepSeek V4 Pro 0813 (max) | 91.7% | #26 of 222 |
 MiniMax | MiniMax-M3 | 90.9% | #31 of 222 |
 Meta AI | Muse Spark | 89.8% | #44 of 222 |
 NVIDIA | nemotron-3-ultra | 85.3% | #65 of 222 |
 ByteDance | ByteDance-Seed/Seed-OSS-36B-Instruct | 71.5% | #103 of 222 |
Every result
222 models, each once at its best configuration, highest score first. Models with the same score share a rank.
| Rank | Model | Maker | Score |
|---|---|---|---|
| #1 | GPT-6 Astra (max) | OpenAI | 95.8% |
| #2 | Claude Sonnet 5.5 (max) | Anthropic | 95.6% |
| #3 | Gemini 3.8 Flash (high) | Google | 95.4% |
| #3 | GPT-6.1 Sol (max) | OpenAI | 95.4% |
| #5 | Gemini 3.7 Flash (high) | Google | 94.8% |
| #6 | GPT-5.4 Pro (xhigh) | OpenAI | 94.6% |
| #7 | Gemini 3.1 Pro (high) | Google | 94.4% |
| #8 | GPT-6 Sol (max) | OpenAI | 94.3% |
| #9 | Gemini 3.6 Flash (high) | Google | 94.1% |
| #10 | Gemini 3.1 Pro Preview | Google | 94.1% |
| #11 | GPT-5.5 Pre-release (xhigh) | OpenAI | 94.0% |
| #11 | Grok 4.6 (high) | xAI | 94.0% |
| #13 | GPT-5.5 Pro Pre-release (xhigh) | OpenAI | 93.9% |
| #14 | Claude Opus 5 (max) | Anthropic | 93.9% |
| #15 | GPT-5.6 Sol (max) | OpenAI | 93.5% |
| #16 | Grok 4.5 (high) | xAI | 93.4% |
| #17 | GPT-5.6 Terra (max) | OpenAI | 93.3% |
| #18 | GPT-5.4 (xhigh) | OpenAI | 93.3% |
| #19 | Kimi K3 (max) | Moonshot AI | 93.1% |
| #20 | Gemini 3.5 Flash (high) | Google | 92.8% |
| #21 | Grok 4.7 (xhigh) | xAI | 92.7% |
| #22 | Qwen3.8 Max (xhigh) | Alibaba | 92.7% |
| #23 | Gemini 3 Pro Preview | Google | 92.6% |
| #24 | Qwen3.8 Max 0902 (xhigh) | Alibaba | 92.3% |
| #25 | GLM-5.2 (max) | Z.ai | 91.9% |
How it is scored
Best published score for each exact model configuration, as evaluated by Epoch AI. A suffix such as _high names the reasoning effort used. Standard errors are shown where the source reports them.
Results from Epoch AI (CC BY 4.0), as published. Model names link to their pages where the site can match them; makers come from the model’s page, or else from the publisher.