MedProbeBench (Deep Research Agents) - Evidence Verification: 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. Evidence verification combining claim recall, search effectiveness (60% reference recall and 40% capped reference quantity), and factual consistency of cited claims, times 100; higher is better.
Metric: Evidence verification (x100): task-weighted composite of claim recall, search effectiveness (60% reference recall plus 40% capped reference quantity) and factual consistency of cited claims, 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 | 57 | |
| 2 | MiroThinker-v1.5 | 55.4 | |
| 3 | Sonar Deep Research | 53.6 | |
| 4 | O4 Mini Deep Research | 50.2 | |
| 5 | AgentScope + GPT-4.1 | 48.4 | |
| 6 | Kimi Agent + Kimi-K2.5 | 47.4 | |
| 7 | Tongyi DeepResearch 30B A3B | 26.1 |
Interactive version: theaggregate.ai/benchmark?slug=medprobebench-deep-research-agents-evidence-verification · How It Works · Data refreshed daily, snapshot 2026-11-21.