List your company

Tool-calling for Clinical Question Answering on AgentBench FHIR (val)

76.7ScoreQWEN3-8B + RL
8.47626.18843.961.612May 13, 2026

Evaluation results

MethodLinks
QWEN3-8B + RLRL Recipe=GRPO + clip-higher + DAPO-filter, rollouts=52026.0576.779.7
GLM-52026.0558.7—
GEMINI 3 FLASH2026.0551.9—
O4-MINI2026.0546.9—
QWEN3-32Brollouts=52026.0539.368.6
QWEN3-14Brollouts=52026.0532.553.8
QWEN3-8Brollouts=52026.0519.747.2
QWEN3-4Brollouts=52026.0511.129.5