AutoResearchBench - Deep Research: leaderboard
Deep Research on AutoResearchBench (2026): a ReAct agent with a scientific literature search tool must track down the single paper that satisfies a progressively specified, multi-step query, within 30 turns. Accuracy in percent; higher is better.
Metric: Accuracy (%) of pinpointing the single target paper on the Deep Research queries (progressive multi-step probing over full-text scientific literature), a ReAct agent with the DeepXiv scientific-literature search tool, at most 30 turns; higher is better. Source: arxiv.org. Saturation forecast: Around August 2028. 13 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Opus 4.6 | 9.39 | #113 | |
| 2 | Gemini 3.1 Pro (Preview) | 7.93 | #99 | |
| 3 | GPT-5.4 | 7.44 | #135 | |
| 4 | Qwen 3.5 397B A17B | 6.97 | #264 | |
| 5 | Claude Sonnet 4.6 | 6.96 | #173 | |
| 6 | Seed 2.0 Pro | 6.8 | #222 | |
| 7 | Kimi K2.5 | 4.69 | #266 | |
| 8 | DeepSeek V3.2 | 4.21 | #419 | |
| 9 | Qwen 3.5 122B A10B | 3.88 | #346 | |
| 10 | Qwen 3 Max | 3.24 | #415 | |
| 11 | MiniMax-M2.5 | 2.91 | #600 | |
| 12 | Gemini 3 Flash | Non-reasoning | 2.75 | #175 |
| 13 | Qwen 3.5 35B A3B | 1.94 | #492 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=autoresearchbench-deep-research · How It Works · Data refreshed daily, snapshot 2026-11-21.