DeepResearch Bench II - Presentation: leaderboard
The presentation rubrics of DeepResearch Bench II (Li et al., 2026): structural and formatting requirements of the task and of a clear, verifiable report. Share of these rubrics a deep research report satisfies over 132 tasks, judged by Gemini-2.5-Pro; higher is better.
Metric: Rubric pass rate (%; presentation 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 | Gemini 3 Pro Deep Research | 91.85 | |
| 2 | Grok Deep Search | 91.42 | |
| 3 | Gemini 2.5 Pro Deep Research | 90.24 | |
| 4 | o3 Deep Research | 89.16 | |
| 5 | Tongyi-DeepResearch-30B-A3B | 86.13 | |
| 6 | Doubao Deep Research | 83.51 | |
| 7 | Perplexity Deep Research | 79.34 | |
| 8 | Qwen3-Max Deep Research | 74.59 |
Interactive version: theaggregate.ai/benchmark?slug=deepresearch-bench-ii-presentation · How It Works · Data refreshed daily, snapshot 2026-11-12.