MedProbeBench (Deep Research Agents): leaderboard
Expert-level medical guideline writing by deep research agents on MedProbeBench (2026): each system writes a complete clinical guideline for 50 tumor entities from WHO classification volumes, compared with the expert guideline. Overall score combining holistic quality and claim-level evidence verification, times 100; higher is better.
Metric: Overall score (x100): task-weighted composite of holistic quality and fine-grained evidence verification, for deep research agents on MedProbeBench: each system writes a complete clinical guideline for 50 WHO-classification tumor entities (one run per task), compared with the expert guideline; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 7 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MiroThinker-v1.5-pro | 63.1 | |
| 2 | MiroThinker-v1.5 | 61.4 | |
| 3 | Sonar Deep Research | 57.6 | |
| 4 | O4 Mini Deep Research | 56.1 | |
| 5 | AgentScope + GPT-4.1 | 55.3 | |
| 6 | Kimi Agent + Kimi-K2.5 | 54 | |
| 7 | Tongyi DeepResearch 30B A3B | 34.8 |
Interactive version: theaggregate.ai/benchmark?slug=medprobebench-deep-research-agents · How It Works · Data refreshed daily, snapshot 2026-11-21.