EduResearchBench: leaderboard
Educational research writing benchmark (EduResearchBench, Yue et al., East China Normal University, 2026) that splits the research lifecycle into 24 atomic tasks in six modules (topic recommendation, quantitative and qualitative research write-ups, policy analysis, theory and survey writing, peer review), each posed over supplied literature in Chinese or English, with prompts synthesized by GPT-5 and Gemini 3 Flash from expert templates. Doubao-seed-1.6 and DeepSeek-V3.2 each write a critique and a 1-5 score, and the two scores are averaged; higher is better.
Metric: Judge score (1-5; overall over 24 atomic research-writing tasks in six modules, mean of Doubao-seed-1.6 and DeepSeek-V3.2 critique-then-score ratings). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 3.63 | #163 |
| 2 | Gemini 3 Flash | 3.34 | #159 |
| 3 | Qwen 2.5 72B Instruct | 2.9 | #1093 |
| 4 | Llama 3 70B Instruct | 2.46 | #1684 |
| 5 | LongWriter-glm4-9B | 2.34 | #2419 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=eduresearchbench · How It Works · Data refreshed daily, snapshot 2026-11-12.