C-Eval
General QA · 2023-05-15
C-Eval is a Chinese evaluation suite of four-option multiple-choice exam questions spanning 52 disciplines at middle-school, high-school, college and professional levels. Scores are accuracy; sources report either the validation split or the test split, so each row's notes record the split and shot count where the source states them.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K2.5 | 94.0 |
| Qwen3 Max Thinking (2026-01-23) | 93.7 |
| Gemini 3 Pro Preview | 93.4 |
| Qwen3.6 Plus (2026-04-02) | 93.3 |
| DeepSeek-V4-Pro-Base | 93.1 |
| Qwen3.5 397B A17B | 93.0 |
| DeepSeek-V3.2 | 92.9 |
| DeepSeek-R1-Zero | 92.8 |
| GLM-5 | 92.8 |
| Kimi K2 Base | 92.5 |
| Qwen3.5 Plus (2026-02-15) | 92.3 |
| Opus 4.5 | 92.2 |
| DeepSeek-V4-Flash-Base | 92.1 |
| Qwen3-235B-A22B | 92.1 |
| Qwen3.5-Omni-Plus | 92.0 |
| Qwen3.5 122B A10B | 91.9 |
| DeepSeek-R1 | 91.8 |
| Qwen3.6 27B | 91.4 |
| Qwen2-72B | 91.0 |
| GPT-5.2 | 90.5 |
| Qwen3.5-27B | 90.5 |
| Qwen3.5 35B A3B | 90.2 |
| Qwen3.6 35B A3B | 90.0 |
| Qwen3-Next-80B-A3B-Thinking | 89.7 |
| Qwen1.5 110B | 89.1 |
| QwQ-32B | 88.4 |
| Qwen3.5-9B | 88.2 |
| Qwen2-57B-A14B | 87.7 |
| Qwen3 30B A3B Thinking 2507 | 87.4 |
| Qwen3 32B | 87.3 |
| Qwen3-30B-A3B | 86.6 |
| DeepSeek-V3 | 86.5 |
| Qwen3 14B | 86.2 |
| Qwen3.5-Omni-Flash | 86.0 |
| o1 (2024-12-17) | 85.5 |
| Qwen3.5-4B | 85.1 |
| Qwen2 72B Instruct | 83.8 |
| Qwen1.5 32B | 83.5 |
| Llama 4 Maverick Instruct | 83.5 |
| Qwen3 8B | 83.4 |