EpiBench (Research Workflows) - Final-Turn Accuracy: leaderboard
Metric: Final-turn accuracy (%): the multi-evidence fusion turn answered from memory, over the 102 EpiBench episodes (39 easy, 63 hard), each a multi-turn research workflow in which the agent searches papers, extracts figures and tables with tools and answers objective questions, the final turn from stored evidence only; answers judged by GPT-5.2; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 | 59.41 |
| 2 | Kimi K2.5 | 57.58 |
| 3 | GPT-5 Mini | 54.9 |
| 4 | Gemini 2.5 Pro | 52.94 |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 51.96 |
| 6 | Grok 4.1 | 50 |
| 7 | Qwen 3 VL 235B A22B Instruct | 46.08 |
| 8 | GLM-4.5V | 38.24 |
Interactive version: theaggregate.ai/benchmark?slug=epibench-research-workflows-final-turn-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-07.