MedProbeBench (Deep Research Agents) - Holistic Quality: 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. Holistic quality: comprehensiveness, insight depth, standards adherence and clinical utility rated by an LLM judge, times 100; higher is better.

Metric: Holistic quality (x100): task-weighted mean of comprehensiveness, insight depth, standards adherence and clinical utility, each rated 0 to 10 by a GPT-4.1 judge against the expert guideline, 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

#ModelScoreOverall rank
1MiroThinker-v1.5-pro77.2
2MiroThinker-v1.575.4
3AgentScope + GPT-4.171.4
4O4 Mini Deep Research69.9
5Kimi Agent + Kimi-K2.569.5
6Sonar Deep Research67.2
7Tongyi DeepResearch 30B A3B55.1

Interactive version: theaggregate.ai/benchmark?slug=medprobebench-deep-research-agents-holistic-quality · How It Works · Data refreshed daily, snapshot 2026-11-21.