大模型排行榜
Arena评测同步于
评分说明与数据来源
综合榜采用 Epoch 能力指数(ECI)。各分类使用页面标注的单项评测,展示该评测的原始百分比分,不合并不同榜单的分数。
评测有多个运行记录时,优先取日期最近的一次。模型详情列出实际参测配置。综合榜展示评测覆盖项数和 ECI 的 90% 可信区间。
考查物理、化学和生物领域的研究生级科学推理。
GPQA Diamond ↗科研榜
232 个模型 · 第 1 / 5 页
| 01 | GPT-6.2 AstraOpenAI · 2026-11-17 | 96.84% |
|---|---|---|
| 02 | Gemini 4 HydrogenGoogle DeepMind · 2026-11-12 | 96.78% |
| 03 | Gemini 4 CarbonGoogle DeepMind · 2026-11-10 | 96.53% |
| 04 | Gemini 4 ArgonGoogle DeepMind · 2026-10-13 | 96.4% |
| 05 | Gemini 4 NeonGoogle DeepMind · 2026-11-18 | 96.28% |
| 06 | Claude Fable 5.2Anthropic · 2026-11-09 | 96.02% |
| 07 | GPT-6 AstraOpenAI · 2026-09-03 | 95.77% |
| 08 | Claude Sonnet 5.5Anthropic · 2026-09-28 | 95.58% |
| 09 | Gemini 3.8 FlashGoogle DeepMind · 2026-09-02 | 95.39% |
| 09 | GPT-6.1 SolOpenAI · 2026-09-29 | 95.39% |
| 11 | Gemini 3.7 FlashGoogle DeepMind · 2026-08-13 | 94.82% |
| 12 | GPT-5.4 ProOpenAI · 2026-03-05 | 94.6% |
| 13 | Gemini 3.1 ProGoogle DeepMind · 2026-02-19 | 94.44% |
| 14 | GPT-6.1 TerraOpenAI · 2026-10-26 | 94.32% |
| 14 | GPT-6 SolOpenAI · 2026-09-22 | 94.32% |
| 16 | Kimi K3.1 ProMoonshot · 2026-10-23 · 开放权重 | 94.19% |
| 17 | Claude Opus 5.5Anthropic · 2026-09-22 | 94.07% |
| 18 | GPT-5.5 ProOpenAI · 2026-04-23 | 93.92% |
| 19 | Claude Fable 5Anthropic · 2026-06-09 | 93.88% |
| 20 | Grok 4.5xAI · 2026-07-08 | 93.43% |
| 21 | DeepSeek V4.1 Flash 1017DeepSeek · 2026-10-17 · 开放权重 | 93.18% |
| 21 | Grok 4.6xAI · 2026-08-12 | 93.18% |
| 23 | Doubao Seed Evolving (Nov 2026)ByteDance · 2026-11-02 | 93.12% |
| 24 | Claude Opus 5Anthropic · 2026-07-24 | 92.93% |
| 25 | Grok 4.7xAI · 2026-09-21 | 92.74% |
| 26 | Qwen 3.8 MaxAlibaba · 2026-08-02 | 92.68% |
| 27 | Gemini 3 ProGoogle DeepMind · 2025-11-18 | 92.61% |
| 28 | Qwen3.8 Max (0902)Alibaba · 2026-09-01 | 92.3% |
| 29 | DeepSeek V4 Pro 0813DeepSeek · 2026-08-13 · 开放权重 | 91.67% |
| 30 | DeepSeek V4 Flash 0731DeepSeek · 2026-07-31 · 开放权重 | 91.04% |
| 31 | GLM-5.3Z.ai (Zhipu AI) · 2026-08-14 · 开放权重 | 90.91% |
| 31 | Qwen3.7-MaxAlibaba · 2026-05-19 | 90.91% |
| 33 | Kimi K2.6Moonshot · 2026-04-20 · 开放权重 | 90.78% |
| 34 | GPT-6 LunaOpenAI · 2026-09-22 | 90.47% |
| 35 | GLM-5.3-FlashZ.ai (Zhipu AI) · 2026-08-20 · 开放权重 | 90.15% |
| 35 | Mistral Large 4Mistral AI · 2026-10-29 · 开放权重 | 90.15% |
| 37 | GLM-5.1Z.ai (Zhipu AI) · 2026-04-07 · 开放权重 | 89.9% |
| 38 | DeepSeek V4.1 FlashDeepSeek · 2026-09-09 · 开放权重 | 89.84% |
| 39 | Muse SparkMeta AI · 2026-04-08 | 89.8% |
| 40 | Claude Haiku 5.5Anthropic · 2026-10-07 | 89.65% |
| 41 | Doubao Seed 2.1 TurboByteDance · 2026-06-23 | 89.58% |
| 42 | Gemini 3 FlashGoogle DeepMind · 2025-12-17 | 89.39% |
| 43 | Doubao Seed 2.1 ProByteDance · 2026-06-23 | 89.33% |
| 43 | Grok 4.20xAI · 2026-02-17 | 89.33% |
| 45 | Gemini 3.5 FlashGoogle DeepMind · 2026-05-19 | 88.89% |
| 45 | GPT-5.4OpenAI · 2026-03-05 | 88.89% |
| 47 | Grok 4.3 BetaxAI · 2026-04-17 | 88.83% |
| 48 | Muse Spark OpenMeta AI · 2026-10-15 · 开放权重 | 88.57% |
| 49 | Inkling-SmallThinking Machines · 2026-07-15 · 开放权重 | 88.51% |
| 50 | Claude Opus 4.6Anthropic · 2026-02-05 | 88.38% |