Benchmark
GPQA
reasoning
general
text
A challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Questions are Google-proof and extremely difficult, with PhD experts reaching 65% accuracy.
语言EN
满分1
参评模型100
模型排名
| 名次 | 模型 | 机构 | 分数 | 来源 |
|---|---|---|---|---|
| 1 | Grok-4 Heavy | xAI | 88.4 | 来源 ↗ |
| 2 | Grok-4 | xAI | 87.5 | 来源 ↗ |
| 3 | Nemotron 3 Ultra 550B A55B | NVIDIA | 87.0 | 来源 ↗ |
| 4 | Gemini 2.5 Pro Preview 06-05 | 86.4 | 来源 ↗ | |
| 5 | GPT-5 | OpenAI | 85.7 | 来源 ↗ |
| 6 | Grok 4 Fast | xAI | 85.7 | 来源 ↗ |
| 7 | Claude 3.7 Sonnet | Anthropic | 84.8 | 来源 ↗ |
| 8 | Grok-3 | xAI | 84.6 | 来源 ↗ |
| 9 | Grok-3 Mini | xAI | 84.0 | 来源 ↗ |
| 10 | Claude Sonnet 4.5 | Anthropic | 83.4 | 来源 ↗ |
| 11 | o3 | OpenAI | 83.3 | 来源 ↗ |
| 12 | Gemini 2.5 Pro | 83.0 | 来源 ↗ | |
| 13 | Gemini 2.5 Flash | 82.8 | 来源 ↗ | |
| 14 | GPT-5 mini | OpenAI | 82.3 | 来源 ↗ |
| 15 | o4-mini | OpenAI | 81.4 | 来源 ↗ |
| 16 | Qwen3-235B-A22B-Thinking-2507 | Alibaba Cloud / Qwen Team | 81.1 | 来源 ↗ |
| 17 | DeepSeek-R1-0528 | DeepSeek | 81.0 | 来源 ↗ |
| 18 | GLM-4.6 | Zhipu AI | 81.0 | 来源 ↗ |
| 19 | Claude Opus 4.1 | Anthropic | 80.9 | 来源 ↗ |
| 20 | GPT OSS 120B | OpenAI | 80.1 | 来源 ↗ |
| 21 | DeepSeek-V3.2-Exp | DeepSeek | 79.9 | 来源 ↗ |
| 22 | Claude Opus 4 | Anthropic | 79.6 | 来源 ↗ |
| 23 | GLM-4.5 | Zhipu AI | 79.1 | 来源 ↗ |
| 24 | o1-pro | OpenAI | 79.0 | 来源 ↗ |
| 25 | o1 | OpenAI | 78.0 | 来源 ↗ |
| 26 | Qwen3-235B-A22B-Instruct-2507 | Alibaba Cloud / Qwen Team | 77.5 | 来源 ↗ |
| 27 | o3-mini | OpenAI | 77.2 | 来源 ↗ |
| 28 | Qwen3-Next-80B-A3B-Thinking | Alibaba Cloud / Qwen Team | 77.2 | 来源 ↗ |
| 29 | Llama 3.1 Nemotron Ultra 253B v1 | NVIDIA | 76.0 | 来源 ↗ |
| 30 | Kimi K2 0905 | Moonshot AI | 75.8 | 来源 ↗ |
| 31 | Claude Sonnet 4 | Anthropic | 75.4 | 来源 ↗ |
| 32 | Kimi K2 Instruct | Moonshot AI | 75.1 | 来源 ↗ |
| 33 | Kimi K2-Instruct-0905 | Moonshot AI | 75.1 | 来源 ↗ |
| 34 | GLM-4.5-Air | Zhipu AI | 75.0 | 来源 ↗ |
| 35 | DeepSeek-V3.1 | DeepSeek | 74.9 | 来源 ↗ |
| 36 | Gemini 2.0 Flash Thinking | 74.2 | 来源 ↗ | |
| 37 | DeepSeek R1 Zero | DeepSeek | 73.3 | 来源 ↗ |
| 38 | o1-preview | OpenAI | 73.3 | 来源 ↗ |
| 39 | Claude Haiku 4.5 | Anthropic | 73.0 | 来源 ↗ |
| 40 | Qwen3-Next-80B-A3B-Instruct | Alibaba Cloud / Qwen Team | 72.9 | 来源 ↗ |
| 41 | GPT OSS 20B | OpenAI | 71.5 | 来源 ↗ |
| 42 | GPT-5 nano | OpenAI | 71.2 | 来源 ↗ |
| 43 | Magistral Medium | Mistral AI | 70.8 | 来源 ↗ |
| 44 | GPT-4o | OpenAI | 70.1 | 来源 ↗ |
| 45 | Llama 4 Maverick | Meta | 69.8 | 来源 ↗ |
| 46 | GPT-4.5 | OpenAI | 69.5 | 来源 ↗ |
| 47 | Phi 4 Reasoning Plus | Microsoft | 68.9 | 来源 ↗ |
| 48 | DeepSeek-V3 0324 | DeepSeek | 68.4 | 来源 ↗ |
| 49 | Magistral Small 2506 | Mistral AI | 68.2 | 来源 ↗ |
| 50 | Claude 3.5 Sonnet | Anthropic | 67.2 | 来源 ↗ |
| 51 | Llama-3.3 Nemotron Super 49B v1 | NVIDIA | 66.7 | 来源 ↗ |
| 52 | GPT-4.1 | OpenAI | 66.3 | 来源 ↗ |
| 53 | Phi 4 Reasoning | Microsoft | 65.8 | 来源 ↗ |
| 54 | Qwen3 30B A3B | Alibaba Cloud / Qwen Team | 65.8 | 来源 ↗ |
| 55 | DeepSeek R1 Distill Llama 70B | DeepSeek | 65.2 | 来源 ↗ |
| 56 | QwQ-32B | Alibaba Cloud / Qwen Team | 65.2 | 来源 ↗ |
| 57 | QwQ-32B-Preview | Alibaba Cloud / Qwen Team | 65.2 | 来源 ↗ |
| 58 | GPT-4.1 mini | OpenAI | 65.0 | 来源 ↗ |
| 59 | Gemini 2.5 Flash-Lite | 64.6 | 来源 ↗ | |
| 60 | Nemotron Nano 9B v2 | NVIDIA | 64.0 | 来源 ↗ |
| 61 | DeepSeek R1 Distill Qwen 32B | DeepSeek | 62.1 | 来源 ↗ |
| 62 | Gemini 2.0 Flash | 62.1 | 来源 ↗ | |
| 63 | o1-mini | OpenAI | 60.0 | 来源 ↗ |
| 64 | Claude 3.5 Sonnet | Anthropic | 59.4 | 来源 ↗ |
| 65 | DeepSeek R1 Distill Qwen 14B | DeepSeek | 59.1 | 来源 ↗ |
| 66 | DeepSeek-V3 | DeepSeek | 59.1 | 来源 ↗ |
| 67 | Gemini 1.5 Pro | 59.1 | 来源 ↗ | |
| 68 | Llama 4 Scout | Meta | 57.2 | 来源 ↗ |
| 69 | Phi 4 | Microsoft | 56.1 | 来源 ↗ |
| 70 | Grok-2 | xAI | 56.0 | 来源 ↗ |
| 71 | Llama 3.1 Nemotron Nano 8B V1 | NVIDIA | 54.1 | 来源 ↗ |
| 72 | GPT-4o | OpenAI | 53.6 | 来源 ↗ |
| 73 | Phi 4 Mini Reasoning | Microsoft | 52.0 | 来源 ↗ |
| 74 | Gemini 2.0 Flash-Lite | 51.5 | 来源 ↗ | |
| 75 | Gemini 1.5 Flash | 51.0 | 来源 ↗ | |
| 76 | Grok-2 mini | xAI | 51.0 | 来源 ↗ |
| 77 | Llama 3.1 405B Instruct | Meta | 50.7 | 来源 ↗ |
| 78 | Llama 3.3 70B Instruct | Meta | 50.5 | 来源 ↗ |
| 79 | Claude 3 Opus | Anthropic | 50.4 | 来源 ↗ |
| 80 | GPT-4.1 nano | OpenAI | 50.3 | 来源 ↗ |
| 81 | Qwen2.5 32B Instruct | Alibaba Cloud / Qwen Team | 49.5 | 来源 ↗ |
| 82 | DeepSeek R1 Distill Qwen 7B | DeepSeek | 49.1 | 来源 ↗ |
| 83 | DeepSeek R1 Distill Llama 8B | DeepSeek | 49.0 | 来源 ↗ |
| 84 | Qwen2.5 72B Instruct | Alibaba Cloud / Qwen Team | 49.0 | 来源 ↗ |
| 85 | Kimi K2 Base | Moonshot AI | 48.1 | 来源 ↗ |
| 86 | GPT-4 Turbo | OpenAI | 48.0 | 来源 ↗ |
| 87 | Qwen3 235B A22B | Alibaba Cloud / Qwen Team | 47.5 | 来源 ↗ |
| 88 | Nova Pro | Amazon | 46.9 | 来源 ↗ |
| 89 | Llama 3.2 90B Instruct | Meta | 46.7 | 来源 ↗ |
| 90 | Mistral Small 3.2 24B Instruct | Mistral AI | 46.1 | 来源 ↗ |
| 91 | Qwen2.5 VL 32B Instruct | Alibaba Cloud / Qwen Team | 46.0 | 来源 ↗ |
| 92 | Mistral Small 3.1 24B Instruct | Mistral AI | 46.0 | 来源 ↗ |
| 93 | Qwen2.5 14B Instruct | Alibaba Cloud / Qwen Team | 45.5 | 来源 ↗ |
| 94 | Mistral Small 3 24B Instruct | Mistral AI | 45.3 | 来源 ↗ |
| 95 | Gemma 3 27B | 42.4 | 来源 ↗ | |
| 96 | Qwen2 72B Instruct | Alibaba Cloud / Qwen Team | 42.4 | 来源 ↗ |
| 97 | Nova Lite | Amazon | 42.0 | 来源 ↗ |
| 98 | Llama 3.1 70B Instruct | Meta | 41.7 | 来源 ↗ |
| 99 | Claude 3.5 Haiku | Anthropic | 41.6 | 来源 ↗ |
| 100 | Gemma 3 12B | 40.9 | 来源 ↗ |