C-Eval
Qwen3-Max 93.7% leads since Jan 2026 Fading 13,948 Chinese multiple-choice exam questions across 52 disciplines and four difficulty levels. Accuracy.
Saturated. The maintainers released the full test set with answers on 2025-07-27, so test scores are no longer held out.
In the authors’ words
New NLP benchmarks are urgently needed to align with the rapid development of large language models (LLMs). We present C-Eval, the first comprehensive Chinese evaluation suite designed to assess advanced knowledge and reasoning abilities of foundation models in a Chinese context. C-Eval comprises multiple-choice questions across four difficulty levels: middle school, high school, college, and professional. The questions span 52 diverse disciplines, ranging from humanities to science and engineering. C-Eval is accompanied by C-Eval Hard, a subset of very challenging subjects in C-Eval that requires advanced reasoning abilities to solve. We conduct a comprehensive evaluation of the most advanced LLMs on C-Eval, including both English- and Chinese-oriented models. Results indicate that only GPT-4 could achieve an average accuracy of over 60%, suggesting that there is still significant room for improvement for current LLMs. We anticipate C-Eval will help analyze important strengths and shortcomings of foundation models, and foster their development and growth for Chinese users.
Made by SJTU + Tsinghua + Edinburgh Released May 2023 Exam questions All tasks public Last in a launch post Sep 2026 , dropped after it saturated
Last model added Sep 2026
- Level
- University Undergraduates and entry-level professionals
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Multiple choice
38
Models scored in lab cards
29
Launch posts list it, from 9 labs
25
System cards report it, from 9 labs
Qwen3-Max leads with 93.7% among models released by Sep 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Sep 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- Claude Sonnet 5.5
- and 16 more
- 93.7 Qwen3-Max
- 93.3 Qwen 3.6 Plus
- 93.1
DeepSeek-V4-Pro - 93.0 Qwen3.5 397B-A17B
- 92.1 DeepSeek V4.1 Flash
- 92.0 Qwen3.5-Omni-Plus
- 91.9 Qwen3.5-122B-A10B
- 91.8 DeepSeek-R1
- 91.8
Hunyuan-T1 - 91.7
LongCat-Flash-Omni - 91.5
Ring-1T - 91.4 Qwen3.6 27B
- 90.9 Hunyuan-TurboS
- 90.9
InternVL3.5-241B-A28B - 90.6
ERNIE-4.5-300B-A47B - 90.4 LongCat-Flash-Chat
- 90.0 Qwen 3.6 35B-A3B
- 89.6 Qwen3-235B-A22B
- 89.2
Keye-VL-2.0-30B-A3B - 88.8 GLM-5
- 88.3 Kimi k1.5
- 87.3 Qwen3-32B
- 86.9 GLM-4.5
- 86.6 Qwen3-30B-A3B
- 86.5 DeepSeek-V3
- 86.2 Qwen3-14B
- 84.1 Qwen1.5-72B
- 83.8 Qwen2-72B
- 83.4 Qwen3-8B
- 78.0 DeepSeek-V2 (MoE-236B, May 2024)
- 77.5 Qwen3-4B
- 77.2 Moonlight-16B-A3B
- 71.7 Qwen-14B
- 69.0 ChatGLM3-6B
- 68.1 Qwen3-1.7B
- 65.2 DeepSeek LLM 67B
- 59.7 Qwen-7B
- 50.7 Qwen-1_8B
- 1 Qwen3-Max 93.7%
- 2 Qwen 3.6 Plus 93.3%
- 3 DeepSeek-V4-Pro 93.1%
- 4 Qwen3.5 397B-A17B 93%
- 5 DeepSeek V4.1 Flash 92.1%
- 6 Qwen3.5-Omni-Plus 92%
- 7 Qwen3.5-122B-A10B 91.9%
- 8 DeepSeek-R1 91.8%
- 9 Hunyuan-T1 91.8%
- 10 LongCat-Flash-Omni 91.7%
- 11 Ring-1T 91.5%
- 12 Qwen3.6 27B 91.4%
- 13 Hunyuan-TurboS 90.9%
- 14 InternVL3.5-241B-A28B 90.9%
- 15 ERNIE-4.5-300B-A47B 90.6%
- 16 LongCat-Flash-Chat 90.4%
- 17 Qwen 3.6 35B-A3B 90%
- 18 Qwen3-235B-A22B 89.6%
- 19 Keye-VL-2.0-30B-A3B 89.2%
- 20 GLM-5 88.8%
- 21 Kimi k1.5 88.3%
- 22 Qwen3-32B 87.3%
- 23 GLM-4.5 86.9%
- 24 Qwen3-30B-A3B 86.6%
- 25 DeepSeek-V3 86.5%
- 26 Qwen3-14B 86.2%
- 27 Qwen1.5-72B 84.1%
- 28 Qwen2-72B 83.8%
- 29 Qwen3-8B 83.4%
- 30 DeepSeek-V2 (MoE-236B, May 2024) 78%
- 31 Qwen3-4B 77.5%
- 32 Moonlight-16B-A3B 77.2%
- 33 Qwen-14B 71.7%
- 34 ChatGLM3-6B 69%
- 35 Qwen3-1.7B 68.1%
- 36 DeepSeek LLM 67B 65.2%
- 37 Qwen-7B 59.7%
- 38 Qwen-1_8B 50.7%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Qwen3-Max | 93.7% | Jan 2026 | Model card | |
| 2 | Qwen 3.6 Plus | 93.3% | Apr 2026 | Model card | |
| 3 | DeepSeek-V4-Pro | 93.1% | Apr 2026 | Model card | |
| 4 | Qwen3.5 397B-A17B | 93% | Feb 2026 | Model card | |
| 5 | DeepSeek V4.1 Flash | 92.1% | Sep 2026 | Model card | |
| 6 | Qwen3.5-Omni-Plus | 92% | Mar 2026 | Model card | |
| 7 | Qwen3.5-122B-A10B | 91.9% | Feb 2026 | Model card | |
| 8 | DeepSeek-R1 | 91.8% | Jan 2025 | Model card | |
| 9 | Hunyuan-T1 | 91.8% | Mar 2025 | Model card | |
| 10 | LongCat-Flash-Omni | 91.7% | Oct 2025 | Model card | |
| 11 | Ring-1T | 91.5% | Oct 2025 | Model card | |
| 12 | Qwen3.6 27B | 91.4% | Apr 2026 | Model card | |
| 13 | Hunyuan-TurboS | 90.9% | Feb 2025 | Model card | |
| 14 | InternVL3.5-241B-A28B | 90.9% | Aug 2025 | Model card | |
| 15 | ERNIE-4.5-300B-A47B | 90.6% | Jun 2025 | Model card | |
| 16 | LongCat-Flash-Chat | 90.4% | Aug 2025 | Model card | |
| 17 | Qwen 3.6 35B-A3B | 90% | Apr 2026 | Model card | |
| 18 | Qwen3-235B-A22B | 89.6% | Apr 2025 | Model card | |
| 19 | Keye-VL-2.0-30B-A3B | 89.2% | May 2026 | Model card | |
| 20 | GLM-5 | 88.8% | Feb 2026 | Model card | |
| 21 | Kimi k1.5 | 88.3% | Jan 2025 | Model card | |
| 22 | Qwen3-32B | 87.3% | May 2025 | Model card | |
| 23 | GLM-4.5 | 86.9% | Aug 2025 | Model card | |
| 24 | Qwen3-30B-A3B | 86.6% | May 2025 | Model card | |
| 25 | DeepSeek-V3 | 86.5% | Dec 2024 | Model card | |
| 26 | Qwen3-14B | 86.2% | May 2025 | Model card | |
| 27 | Qwen1.5-72B | 84.1% | Feb 2024 | Model card | |
| 28 | Qwen2-72B | 83.8% | Jun 2024 | Model card | |
| 29 | Qwen3-8B | 83.4% | May 2025 | Model card | |
| 30 | DeepSeek-V2 (MoE-236B, May 2024) | 78% | May 2024 | Model card | |
| 31 | Qwen3-4B | 77.5% | May 2025 | Model card | |
| 32 | Moonlight-16B-A3B | 77.2% | Feb 2025 | Model card | |
| 33 | Qwen-14B | 71.7% | Sep 2023 | Model card | |
| 34 | ChatGLM3-6B | 69% | Oct 2023 | Model card | |
| 35 | Qwen3-1.7B | 68.1% | May 2025 | Model card | |
| 36 | DeepSeek LLM 67B | 65.2% | Nov 2023 | Model card | |
| 37 | Qwen-7B | 59.7% | Sep 2023 | Model card | |
| 38 | Qwen-1_8B | 50.7% | Sep 2023 | Model card |
5 models have held number one since its release in May 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Qwen DeepSeek
The best score passed 85% in Dec 2024
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Reported by the lab in its model card
Best so far
Labs headline it less often now, last in Sep 2026
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 10 Sep 2026 | DeepSeek V4.1 Flash | 92.1 | C-Eval (EM) |
| 25 May 2026 | Keye-VL-2.0-30B-A3B | 89.2 | C-Eval |
| 24 Apr 2026 | DeepSeek-V4-Pro | 93.1 | C-Eval (EM) |
| 22 Apr 2026 | Qwen3.6 27B | 91.4 | C-Eval |
| 15 Apr 2026 | Qwen 3.6 35B-A3B | 90 | C-Eval |
| 2 Apr 2026 | Qwen 3.6 Plus | 93.3 | C-Eval |
| 30 Mar 2026 | Qwen3.5-Omni-Plus | 92 | C-Eval |
| 24 Feb 2026 | Qwen3.5-122B-A10B | 91.9 | C-Eval |
| 16 Feb 2026 | Qwen3.5 397B-A17B | 93 | C-Eval |
| 23 Jan 2026 | Qwen3-Max | 93.7 | C-Eval |
| 31 Oct 2025 | LongCat-Flash-Omni | 91.7 | CEval(acc) |
| 29 Aug 2025 | LongCat-Flash-Chat | 90.4 | CEval(acc) |
| 25 Aug 2025 | InternVL3.5-241B-A28B | 90.9 | C-Eval |
| 30 Jun 2025 | ERNIE-4.5-300B-A47B | 90.6 | C-Eval |
| 29 Apr 2025 | Qwen3-235B-A22B | 89.6 | C-Eval |
| 21 Mar 2025 | Hunyuan-T1 | 91.8 | CEval |
| 16 Mar 2025 | ERNIE 4.5 | C-Eval | |
| 27 Feb 2025 | Hunyuan-TurboS | 90.9 | C-Eval |
| 22 Feb 2025 | Moonlight-16B-A3B | 77.2 | C-Eval |
| 28 Jan 2025 | Qwen2.5-Max | C-Eval | |
| 20 Jan 2025 | Kimi k1.5 | 88.3 | C-Eval (EM), short-CoT |
| 20 Jan 2025 | DeepSeek-R1 | 91.8 | C-Eval (EM) |
| 26 Dec 2024 | DeepSeek-V3 | 86.5 | C-Eval (EM) |
| 7 Jun 2024 | Qwen2-72B | 83.8 | C-Eval |
| 7 May 2024 | DeepSeek-V2 (MoE-236B, May 2024) | 78 | C-Eval |
| 5 Feb 2024 | Qwen1.5-72B | 84.1 | C-Eval |
| 29 Nov 2023 | DeepSeek LLM 67B | 65.2 | C-Eval |
| 27 Oct 2023 | ChatGLM3-6B | 69 | C-Eval |
| 25 Sep 2023 | Qwen-14B | 71.7 | C-Eval (val, 5-shot) |
25 system cards report it, from 9 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 9 Jun 2026 | Keye-VL-2.0-30B-A3B technical report | 89.2 | C-Eval (Figure 5, printed value) | Capabilities |
| 17 Apr 2026 | Qwen3.5-Omni-Plus technical report | 92 | C-Eval (Table 4) | Capabilities |
| 17 Feb 2026 | GLM-5 technical report | 88.8 | GLM-5-Base, C-Eval (EM) (Table 11) | Capabilities |
| 31 Oct 2025 | LongCat-Flash-Omni technical report | 91.7 | CEval (acc) (Table 13) | Capabilities |
| 21 Oct 2025 | Ring-1T technical report | 91.5 | C-Eval (Table 1) | Capabilities |
| 1 Sep 2025 | LongCat-Flash-Chat technical report | 90.4 | CEval (acc) (Table 3) | Capabilities |
| 25 Aug 2025 | InternVL3.5-241B-A28B technical report | 90.9 | C-Eval (Table 2, Table 14) | Capabilities |
| 8 Aug 2025 | GLM-4.5 technical report | 86.9 | GLM-4.5-Base, C-Eval (EM) (Table 2) | Capabilities |
| 29 Jun 2025 | ERNIE-4.5-300B-A47B technical report | 90.6 | C-Eval (Table 5) | Capabilities |
| 14 May 2025 | Qwen3-8B technical report | 83.4 | C-Eval, thinking mode (Table 17). Non-thinking 77.9 (Table 18) | Capabilities |
| 14 May 2025 | Qwen3-4B technical report | 77.5 | C-Eval, thinking mode (Table 17). Non-thinking 72.2 (Table 18) | Capabilities |
| 14 May 2025 | Qwen3-32B technical report | 87.3 | C-Eval, thinking mode (Table 13). Non-thinking 83.3 (Table 14) | Capabilities |
| 14 May 2025 | Qwen3-30B-A3B technical report | 86.6 | C-Eval, thinking mode (Table 15). Non-thinking 82.9 (Table 16) | Capabilities |
| 14 May 2025 | Qwen3-235B-A22B technical report | 89.6 | C-Eval, thinking mode (Table 11). Non-thinking 86.1 (Table 12) | Capabilities |
| 14 May 2025 | Qwen3-14B technical report | 86.2 | C-Eval, thinking mode (Table 15). Non-thinking 81.0 (Table 16) | Capabilities |
| 14 May 2025 | Qwen3-1.7B technical report | 68.1 | C-Eval, thinking mode (Table 19). Non-thinking 61.0 (Table 20) | Capabilities |
| 20 Jan 2025 | Kimi k1.5 technical report | 88.3 | C-Eval (EM), short-CoT (Table 3) | Capabilities |
| 20 Jan 2025 | DeepSeek-R1 technical report | 91.8 | C-Eval (EM), DeepSeek-R1 (Table 8, same as final column of Table 3) | Capabilities |
| 26 Dec 2024 | DeepSeek-V3 technical report | 86.5 | C-Eval (EM), DeepSeek-V3 (Table 6) | Capabilities |
| 18 Jun 2024 | ChatGLM3-6B technical report | 69 | ChatGLM3-6B-Base, C-Eval (Table 1) | Capabilities |
| 7 May 2024 | DeepSeek-V2 (MoE-236B, May 2024) technical report | 78 | C-Eval 5-shot, DeepSeek-V2 Chat (RL) (Table 3) | Capabilities |
| 5 Jan 2024 | DeepSeek LLM 67B technical report | 65.2 | C-Eval 0-shot, DeepSeek 67B Chat (Table 6) | Capabilities |
| 28 Sep 2023 | Qwen-7B technical report | 59.7 | C-Eval 0-shot, higher of the two settings, 5-shot 59.3 (Table 5) | Capabilities |
| 28 Sep 2023 | Qwen-14B technical report | 71.7 | C-Eval 5-shot, higher of the two settings, 0-shot 69.8 (Table 5) | Capabilities |
| 28 Sep 2023 | Qwen-1_8B technical report | 50.7 | C-Eval 0-shot, higher of the two settings, 5-shot 50.3 (Table 5) | Capabilities |
More in Knowledge
- MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
- SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
- WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
- KMMLU-Pro In use Models answer multiple-choice questions taken from Korean National Professional Licensure exams. It tests professional knowledge used in Korean industry.
Solar Pro 4 79.2% - BIG-Bench Extra Hard In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is a harmonic mean of task accuracies.
Gemma 4 31B IT 74.4% - HLE-Verified In use Verified and revised version of Humanity's Last Exam (1,811 items) with uncertain items removed.
Seed2.0 Pro 73.6%