AutoResearchBench - Deep Research: leaderboard
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 2029. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 9.39 |
| 2 | Gemini 3.1 Pro (Preview) | 7.93 |
| 3 | GPT-5.4 | 7.44 |
| 4 | Qwen 3.5 397B A17B | 6.97 |
| 5 | Claude Sonnet 4.6 | 6.96 |
| 6 | Seed 2.0 Pro | 6.8 |
| 7 | Kimi K2.5 | 4.69 |
| 8 | DeepSeek V3.2 | 4.21 |
| 9 | Qwen 3.5 122B A10B | 3.88 |
| 10 | Qwen 3 Max | 3.24 |
| 11 | MiniMax-M2.5 | 2.91 |
| 12 | Qwen 3.5 35B A3B | 1.94 |
Interactive version: theaggregate.ai/benchmark?slug=autoresearchbench-deep-research · How It Works · Data refreshed daily, snapshot 2026-10-07.