AutoResearchBench - Wide Research: leaderboard

Wide Research on AutoResearchBench (2026): a ReAct agent with a scientific literature search tool must collect every paper that satisfies a set of conditions, within 30 turns. Intersection over union with the gold set, in percent; higher is better.

Metric: Intersection over union (%) between the collected paper set and the gold set of papers satisfying the query conditions on the Wide Research queries, a ReAct agent with the DeepXiv scientific-literature search tool, at most 30 turns; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.1 Pro (Preview)9.31#99
2GPT-5.48.12#135
3Seed 2.0 Pro7.87#222
4DeepSeek V3.27.7#419
5Qwen 3 Max6.89#415
6Gemini 3 FlashNon-reasoning6.61#175
7Claude Opus 4.66.56#113
8Kimi K2.56.23#266
9Claude Sonnet 4.65.83#173
10Qwen 3.5 397B A17B3.83#264
11Qwen 3.5 122B A10B2.76#346
12Qwen 3.5 35B A3B2.71#492
13MiniMax-M2.51.39#600

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=autoresearchbench-wide-research · How It Works · Data refreshed daily, snapshot 2026-11-21.