LLMEval-Fair - History: leaderboard
LLMEval-Fair (Fudan NLP; ACL 2026): generative questions from Chinese undergraduate coursework, exams and graduate entrance exams; each model answers its own random draw of 1,000 questions, scored 0-3 by a GPT-4 Turbo judge. Frozen board as of December 2025. History discipline score on the source's 10-point scale; higher is better.
Metric: Discipline Score (0-10). Source: github.com. Saturation forecast: Estimated already saturated. 59 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Doubao-1.5-Thinking-Pro | 9.7 | #140 |
| 2 | DeepSeek R1 | 9.53 | #262 |
| 3 | DeepSeek V3 | 9.47 | #329 |
| 4 | Gemini 2.5 Pro (Preview) | 9.4 | #136 |
| 5 | DeepSeek V3.2 | 9.37 | #210 |
| 6 | Qwen 3 32B | 9.3 | #449 |
| 7 | GLM-4.6 | 9.3 | #265 |
| 8 | QwQ-32B | 9.27 | #422 |
| 9 | Qwen 3 235B A22B | 9.17 | #323 |
| 10 | Kimi K2 | 9.13 | #252 |
| 11 | GPT-5 | 9.03 | #105 |
| 12 | Claude Sonnet 4.5 (Thinking) | 9 | #148 (Claude Sonnet 4.5) |
| 13 | Gemini 2.5 Flash (Thinking) | 9 | #255 (Gemini 2.5 Flash) |
| 14 | O1 (2024-12-17) | 8.97 | #157 |
| 15 | Claude Sonnet 4.5 | 8.93 | #148 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llmeval-fair-history · How It Works · Data refreshed daily, snapshot 2026-11-05.