CMMLU — Chinese MMLU-style multitask benchmark, base metric
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Kimi K2 Base | Moonshot | 90.9 | 9 | 2026-06-13 |
| 2 | DeepSeek-V4-Pro-Base | DeepSeek | 90.8 | 3 | 2026-06-05 |
| 3 | DeepSeek-V4-Flash-Base | DeepSeek | 90.4 | 3 | 2026-06-05 |
| 4 | Hunyuan Large Inst. | Tencent | 90.4 | 1 | 2026-05-31 |
| 5 | MiMo V2.5 Pro Base | Xiaomi | 90.2 | 3 | 2026-06-05 |
| 6 | MiMo V2.5 Pro | Xiaomi | 90.2 | 1 | 2026-08-23 |
| 7 | Hunyuan Large | Tencent | 90.2 | 1 | 2026-05-31 |
| 8 | Qwen2 Instruct (72B) | Alibaba | 90.1 | 1 | 2026-08-23 |
| 9 | Hy3 preview-Base | Tencent | 89.6 | 2 | 2026-06-05 |
| 10 | Qwen2.5 Instruct 72B | Alibaba | 89.5 | 1 | 2026-05-01 |
| 11 | Step3.5 Flash Base | StepFun | 88.9 | 2 | 2026-06-05 |
| 12 | DeepSeek-V3.2-Exp-Base | DeepSeek | 88.9 | 4 | 2026-06-13 |
| 13 | DeepSeek-V3.1-Base | DeepSeek | 88.8 | 4 | 2026-06-13 |
| 14 | DeepSeek-V3 | DeepSeek | 88.8 | 1 | 2026-05-01 |
| 15 | MiMo V2.5 Base | Xiaomi | 88.2 | 3 | 2026-06-05 |
| 16 | DeepSeek-V3-Base | DeepSeek | 87.9 | 2 | 2026-06-05 |
| 17 | MiMo V2 Flash Base | Xiaomi | 87.4 | 4 | 2026-06-13 |
| 18 | GLM-4.5-Base | Z.ai | 86.5 | 2 | 2026-06-05 |
| 19 | DeepSeek-V2 | DeepSeek | 84 | 2 | 2026-05-31 |
| 20 | LongCat Flash Lite | Meituan | 82.5 | 1 | 2026-08-23 |
| 21 | Moonlight | Moonshot | 78.2 | 2 | 2026-06-12 |
| 22 | Qwen2.5 3B | Alibaba | 75 | 2 | 2026-06-05 |
| 23 | Llama 3.1 Instruct 405B | Meta | 73.7 | 1 | 2026-05-01 |
| 24 | ERNIE 4.5 | Baidu | 39.8 | 1 | 2026-08-23 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.