New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
CMMLU benchmark
AI model leaderboard for the CMMLU benchmark. Compare how large language models score on CMMLU, see the full ranking, and understand what this AI benchmark measures. Qwen3-235B-A22B-Thinking (Jul 2025) currently leads with 90.64. Chinese MMLU — 67-subject Chinese-language multiple-choice exam benchmark.
Leaderboard
| # | Model | Organization | Score | Variant | Source |
|---|---|---|---|---|---|
| #1 | Qwen3-235B-A22B-Thinking (Jul 2025) | Qwen | 90.64 | thinking-cited-ring-1t | official ↗ |
| #2 | TeleChat2-35B | China Telecom | 90.18 | from-hf-readme | official ↗ |
| #3 | Telechat2-115B | China Telecom | 89.94 | from-hf-readme | official ↗ |
| #4 | DeepSeek-V3.1-Terminus | DeepSeek | 89.2 | thinking-cited-ring-1t | official ↗ |
| #5 | Ring-1T | Ant Group | 89.14 | thinking | official ↗ |
| #6 | gemini-2.5-pro | Google DeepMind | 88.83 | cited-ring-1t | official ↗ |
| #7 | Qwen2-57B-A14B | Qwen | 88.5 | base-from-hf-readme | official ↗ |
| #8 | Qwen1.5-110B | Qwen | 87.64 | cited-telechat2 | official ↗ |
| #9 | Qwen2-72B | Qwen | 87.47 | instruct-cited-telechat2 | official ↗ |
| #10 | Pangu Pro MoE | Huawei | 87.1 | instruct-em | official ↗ |
| #11 | GPT-5 | OpenAI | 87.09 | thinking-high-cited-ring-1t | official ↗ |
| #12 | Yi-1.5-34B | 01.AI | 84.8 | cited-qwen2-57b | official ↗ |
| #13 | TeleChat2-7B | China Telecom | 84.6 | from-hf-readme | official ↗ |
| #14 | Qwen2-7B | Qwen | 83.9 | base-from-hf-readme | official ↗ |
| #15 | Qwen1.5-32B | Qwen | 82.3 | cited-qwen2-57b | official ↗ |
| #16 | DeepSeek-V2 (MoE-236B) | DeepSeek | 81.6 | cited-telechat2 | official ↗ |
| #17 | Qwen1.5-7B | Qwen | 73.1 | base-cited-qwen2-7b | official ↗ |
| #18 | TeleChat2-3B | China Telecom | 73 | from-hf-readme | official ↗ |
| #19 | Qwen2-1.5B | Qwen | 70.3 | base-from-hf-readme | official ↗ |
| #20 | Llama 4 Scout | Meta AI | 70 | cited-pangu-pro-moe | official ↗ |
| #21 | Llama 3.1-70B | Meta AI | 69.01 | cited-telechat2 | official ↗ |
| #22 | Qwen2-0.5B | Qwen | 55.1 | base-from-hf-readme | official ↗ |
| #23 | Llama 3-8B | Meta AI | 50.8 | base-cited-qwen2-7b | official ↗ |
Frequently asked questions about CMMLU
What is the CMMLU benchmark?
Chinese MMLU — 67-subject Chinese-language multiple-choice exam benchmark.
How is the CMMLU benchmark scored?
CMMLU is scored using the accuracy metric, where a higher score is better. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.
Which AI model scores highest on CMMLU?
As of the latest reported scores on GenAIList, Qwen3-235B-A22B-Thinking (Jul 2025) achieves the highest result on CMMLU with a score of 90.64.
Is a higher CMMLU score better?
Yes. On CMMLU a higher score indicates better performance, so models near the top of the leaderboard are the strongest.