Open Chinese LLM - CMMLU: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 338 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Smaug-Llama-3-70B-Instruct | 80.7 |
| 2 | Qwen 2 72B | 78.66 |
| 3 | Qwen 2 72B Instruct | 78.4 |
| 4 | Higgs-Llama-3-70B | 75.33 |
| 5 | free-evo-qwen72B-v0.8-re | 73.88 |
| 6 | TW3-JRGL-v2 | 73.73 |
| 7 | Llama 3 70B Synthia V3.5 | 73.46 |
| 8 | Rhea-72B-v0.5 | 73.31 |
| 9 | Meta-LLama-3-Cat-Smaug-LLama-70B | 72.48 |
| 10 | autotrain-llama3-70B-orpo-v1 | 72.42 |
| 11 | dolphin-2.9.1-llama-3-70B | 72.34 |
| 12 | Smaug-72B-v0.1 | 72.32 |
| 13 | Llama 3 70B Japanese Suzume Vector V0.1 | 72.14 |
| 14 | Llama3-TenyxChat-70B | 72.02 |
| 15 | Le_Triomphant-ECE-TW3 | 71.86 |
Interactive version: theaggregate.ai/benchmark?slug=open-chinese-llm-cmmlu · How It Works · Data refreshed daily, snapshot 2026-09-05.