DEEPMED: Building a Medical DeepResearch Agent via Multi-hop Med-Search Data and Turn-Controlled Agentic Training & Inference

TL;DR

DeepMed is a novel medical reasoning framework that enhances multi-hop evidence retrieval, controls tool-use during training and inference, and significantly improves performance across multiple medical benchmarks.

Contribution

It introduces a multi-hop med-search data synthesis, difficulty-aware turn-penalty training, and a hypothesis validation monitor to improve medical reasoning in DR models.

Findings

Achieves 9.79% average improvement on seven benchmarks.

Outperforms larger medical reasoning and DR models.

Effectively manages tool-use and evidence interpretation in medical contexts.

Abstract

Medical reasoning models remain constrained by parametric knowledge and are thus susceptible to forgetting and hallucinations. DeepResearch (DR) models ground outputs in verifiable evidence from tools and perform strongly in general domains, but their direct transfer to medical field yields relatively limited gains. We attribute this to two gaps: task characteristic and tool-use scaling. Medical questions require evidence interpretation in a knowledge-intensive clinical context; while general DR models can retrieve information, they often lack clinical-context reasoning and thus "find it but fail to use it," leaving performance limited by medical abilities. Moreover, in medical scenarios, blindly scaling tool-call can inject noisy context, derailing sensitive medical reasoning and prompting repetitive evidence-seeking along incorrect paths. Therefore, we propose DeepMed. For data, we deploy a multi-hop med-search QA synthesis method supporting the model to apply the DR paradigm in medical contexts. For training, we introduce a difficulty-aware turn-penalty to suppress excessive tool-call growth. For inference, we bring a monitor to help validate hypotheses within a controlled number of steps and avoid context rot. Overall, on seven medical benchmarks, DeepMed improves its base model by 9.79\% on average and outperforms larger medical reasoning and DR models.