Open Chinese LLM - CMMLU: leaderboard
The "CMMLU" subtask of Open Chinese LLM. A task from the Open Chinese LLM Leaderboard, evaluating models on Chinese-language knowledge, reasoning and understanding.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 339 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Smaug-Llama-3-70B-Instruct | 80.7 | #716 |
| 2 | Qwen 2 72B | 78.66 | #521 |
| 3 | Qwen 2 72B Instruct | 78.4 | #576 |
| 4 | free-evo-qwen72B-v0.8-re | 73.88 | #435 |
| 5 | Rhea-72B-v0.5 | 73.31 | #556 |
| 6 | autotrain-llama3-70B-orpo-v1 | 72.42 | #828 |
| 7 | dolphin-2.9.1-llama-3-70B | 72.34 | #722 |
| 8 | Smaug-72B-v0.1 | 72.32 | #611 |
| 9 | Llama 3 70B Japanese Suzume Vector V0.1 | 72.14 | #765 |
| 10 | Llama3-TenyxChat-70B | 72.02 | #735 |
| 11 | Le_Triomphant-ECE-TW3 | 71.86 | #559 |
| 12 | 34B-beta | 71.71 | #903 |
| 13 | Llama 3 70B Fireplace | 71.6 | #833 |
| 14 | bagel-dpo-34B-v0.5 | 71.31 | #962 |
| 15 | Yi-34Bx2-MoE-60B-DPO | 71.11 | #714 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-chinese-llm-cmmlu · How It Works · Data refreshed daily, snapshot 2026-10-20.