Benchmark
开源模型能力榜
general
text
多语言
开源模型能力榜：多维度评测综合成绩。评测分数为事实数据，整理自公开评测体系，仅供学习研究参考。
Top 10
榜单概览
数据说明
分数为 0-100 百分制综合成绩（Elo 榜为相对评分，仅供同榜内比较）。评测分数属事实数据，整理自公开评测体系与厂商公开资料，部分分数为厂商自报口径，不同榜单的评测方法与数据集可能存在差异，请结合多个榜单交叉参考。
Ranking
模型排名
| 名次 | 模型 | 机构 | 主分 | HLE | AIME2025 | MMLU-Pro | LiveCodeBenchV6 | GPQA-Diamond | IFEval |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Gemini 3 Pro Preview | 81.3 | 38.0 | 93.4 | 89.3 | 82.9 | 91.5 | 92.8 | |
| 2 | GLM-5 | Zhipu AI | 79.0 | 28.1 | 95.8 | 85.2 | 86.2 | 85.3 | 93.2 |
| 3 | GPT-5 | OpenAI | 78.8 | 28.5 | 94.1 | 86.2 | 83.2 | 86.4 | 94.6 |
| 4 | Qwen3.5 397B-A17B | Alibaba Cloud / Qwen Team | 78.4 | 27.5 | 92.3 | 87.6 | 83.0 | 88.4 | 91.5 |
| 5 | Kimi K2.5 | Moonshot AI | 78.2 | 28.6 | 91.9 | 86.2 | 80.6 | 88.1 | 93.9 |
| 6 | DeepSeek-V3.2-Speciale | DeepSeek | 78.2 | 28.6 | 96.0 | 85.5 | 80.9 | 86.7 | 91.7 |
| 7 | GLM-4.7 | Zhipu AI | 77.6 | 25.4 | 95.4 | 84.0 | 83.8 | 86.9 | 90.2 |
| 8 | Step 3.5 Flash | StepFun | 76.9 | 21.6 | 95.7 | 83.5 | 83.9 | 83.7 | 93.2 |
| 9 | Kimi K2 Thinking | Moonshot AI | 75.3 | 21.3 | 94.1 | 84.3 | 77.1 | 82.7 | 92.4 |
| 10 | DeepSeek-V3.2 | DeepSeek | 75.3 | 23.2 | 93.0 | 85.8 | 75.4 | 84.6 | 89.7 |
| 11 | Gemini 2.5 Pro | 73.6 | 21.1 | 88.7 | 85.8 | 71.3 | 84.7 | 90.0 | |
| 12 | GLM-4.6 | Zhipu AI | 73.3 | 19.3 | 90.3 | 83.0 | 78.2 | 80.4 | 88.7 |
| 13 | MiniMax-M2.5 | MiniMax | 73.3 | 22.2 | 86.2 | 81.7 | 73.6 | 84.6 | 91.1 |
| 14 | MiMo-V2-Flash | Xiaomi | 73.3 | 20.5 | 92.9 | 83.1 | 71.8 | 82.1 | 89.5 |
| 15 | GPT OSS 120B | OpenAI | 73.2 | 18.3 | 93.4 | 79.7 | 78.4 | 78.9 | 90.2 |
| 16 | GPT-5 mini | OpenAI | 73.2 | 21.5 | 90.9 | 70.2 | 80.8 | 82.7 | 92.8 |
| 17 | o3 | OpenAI | 72.8 | 21.3 | 85.1 | 83.0 | 72.7 | 82.3 | 92.2 |
| 18 | TeleChat-thinking | TeleAI | 71.8 | 20.3 | 79.0 | 84.1 | 71.7 | 81.4 | 94.3 |
| 19 | Qwen3-235B-A22B-Thinking-2507 | Alibaba Cloud / Qwen Team | 71.8 | 18.5 | 90.9 | 83.5 | 70.6 | 79.8 | 87.8 |
| 20 | o4-mini | OpenAI | 71.5 | 14.3 | 91.0 | 82.4 | 71.0 | 77.4 | 93.0 |
| 21 | Qwen3-Next-80B-A3B-Thinking | Alibaba Cloud / Qwen Team | 69.5 | 13.5 | 89.0 | 82.0 | 66.3 | 77.0 | 89.5 |
| 22 | MiniMax-M2 | MiniMax | 69.5 | 13.4 | 79.1 | 81.6 | 74.0 | 78.7 | 90.2 |
| 23 | GLM-4.5 | Zhipu AI | 69.2 | 16.9 | 85.8 | 82.7 | 65.0 | 79.5 | 85.4 |
| 24 | DeepSeek-R1-0528 | DeepSeek | 68.1 | 14.4 | 89.0 | 83.5 | 61.0 | 80.6 | 80.0 |
| 25 | GPT-5 nano | OpenAI | 67.2 | 11.0 | 85.4 | 75.0 | 74.0 | 67.8 | 90.2 |
| 26 | JT-Think | China Mobile | 67.2 | 14.4 | 65.9 | 85.6 | 64.9 | 77.7 | 94.6 |
| 27 | Seed-OSS-36B-Instruct | ByteDance | 66.5 | 9.5 | 86.0 | 80.6 | 67.3 | 71.5 | 83.7 |
| 28 | GPT OSS 20B | OpenAI | 66.4 | 11.6 | 87.9 | 72.8 | 68.4 | 68.9 | 88.9 |
| 29 | Qwen3-30B-A3B-Thinking-2507 | Alibaba Cloud / Qwen Team | 66.2 | 11.7 | 86.8 | 79.5 | 59.4 | 70.1 | 89.7 |
| 30 | doubao-seed-1.6-thinking-250615 | ByteDance | 66.0 | 13.5 | 82.8 | 83.1 | 61.8 | 75.9 | 78.9 |
| 31 | Qwen3-Next-80B-A3B-Instruct | Alibaba Cloud / Qwen Team | 62.5 | 8.0 | 69.2 | 81.3 | 54.8 | 74.1 | 87.6 |
| 32 | Finix-P1-32B (Thinking) | Ant Group | 62.0 | 6.9 | 70.2 | 82.7 | 54.9 | 68.7 | 88.7 |
| 33 | Claude Sonnet 4 | Anthropic | 61.8 | 8.7 | 68.7 | 83.0 | 47.5 | 74.6 | 88.3 |
| 34 | Qwen3 32B | Alibaba Cloud / Qwen Team | 61.3 | 8.5 | 70.3 | 78.0 | 57.6 | 67.3 | 86.0 |
| 35 | Qwen3-235B-A22B-Instruct-2507 | Alibaba Cloud / Qwen Team | 61.3 | 12.3 | 69.5 | 79.2 | 43.0 | 75.5 | 88.3 |
| 36 | Qwen3-4B-Thinking-2507 | Alibaba Cloud / Qwen Team | 60.6 | 6.0 | 80.0 | 72.8 | 51.6 | 64.7 | 88.5 |
| 37 | LongCat-Flash-Chat | Meituan | 59.8 | 8.5 | 61.0 | 81.0 | 48.2 | 70.1 | 90.2 |
| 38 | MiniCPM-SALA | OpenBMB | 58.7 | 9.3 | 81.0 | 72.0 | 54.7 | 58.2 | 76.9 |
| 39 | Kimi K2 Instruct | Moonshot AI | 57.0 | 6.7 | 49.1 | 78.4 | 46.8 | 72.7 | 88.5 |
| 40 | ERNIE-4.5-21B-A3B-Thinking | Baidu | 57.0 | 6.5 | 76.2 | 70.8 | 47.0 | 60.2 | 81.2 |
| 41 | Qwen3-8B (Thinking) | Alibaba Cloud / Qwen Team | 56.5 | 5.5 | 66.2 | 72.1 | 50.1 | 59.7 | 85.6 |
| 42 | Qwen3-30B-A3B-Instruct-2507 | Alibaba Cloud / Qwen Team | 55.3 | 7.7 | 63.8 | 73.9 | 38.7 | 63.6 | 83.9 |
| 43 | ERNIE-4.5-300B-A47B-PT | Baidu | 54.7 | 3.8 | 32.6 | 78.7 | 41.0 | 83.0 | 88.9 |
| 44 | GPT-4.1 | OpenAI | 52.6 | 6.0 | 35.3 | 81.0 | 37.2 | 68.1 | 88.2 |
| 45 | Hunyuan-A13B-Instruct | Tencent | 52.3 | 6.0 | 65.7 | 73.8 | 43.9 | 64.1 | 60.3 |
| 46 | Llama4-Maverick-17B-128E-Instruct | Meta | 48.2 | 5.3 | 15.2 | 79.6 | 33.2 | 69.1 | 87.1 |
| 47 | Qwen3-4B-Instruct-2507 | Alibaba Cloud / Qwen Team | 47.2 | 5.1 | 46.9 | 63.0 | 33.5 | 52.3 | 82.4 |
| 48 | Gemma 3 27B | 42.1 | 4.2 | 22.4 | 67.8 | 30.8 | 46.3 | 81.0 |