Back to DIVE

MedAgentBench

L3 OOD — Specialized Tools TaskPoolSetProtoEnv

Medical EHR (Electronic Health Record) system interaction via FHIR GET/POST endpoints. A stateful HTTP environment requiring clinical reasoning and proper API sequencing for patient data retrieval and medical decision-making.

Tool Pool FHIR GET / POST / Finish
Toolset Uniform
Protocol HTTP (non-OpenAI)
Environment Stateful
Performance (Success Rate %)
Qwen3-8B (base)
38.4
EnvScaler-8B
56.6
DIVE-8B (SFT)
50.2
DIVE-8B (RL)
57.3
Base Best 8B Baseline DIVE (SFT) DIVE (RL)