DeepResearch Bench II - Information Recall: leaderboard
The information-recall rubrics of DeepResearch Bench II (Li et al., 2026): atomic facts and figures from an expert review article that a deep research report on the same question must retrieve and state correctly. Share of these rubrics a report satisfies over 132 tasks, judged by Gemini-2.5-Pro; higher is better.
Metric: Rubric pass rate (%; information-recall rubrics; 132 report tasks in 22 domains; share of expert-derived binary rubrics a report satisfies, Gemini-2.5-Pro judge). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | o3 Deep Research | 39.98 | |
| 2 | Gemini 3 Pro Deep Research | 39.09 | |
| 3 | Gemini 2.5 Pro Deep Research | 34.91 | |
| 4 | Doubao Deep Research | 34.83 | |
| 5 | Qwen3-Max Deep Research | 34.18 | |
| 6 | Grok Deep Search | 33.52 | |
| 7 | Perplexity Deep Research | 33.05 | |
| 8 | Tongyi-DeepResearch-30B-A3B | 22.95 |
Interactive version: theaggregate.ai/benchmark?slug=deepresearch-bench-ii-information-recall · How It Works · Data refreshed daily, snapshot 2026-11-12.