List your company

Post-RL Evaluation on HealthBench Hard

32.1AccuracyMRRG
25.3427.09528.8530.605Jul 2, 2026

Evaluation results

MethodLinks
MRRG2026.0732.1
SVRGSub-reward (SR)=true2026.0728.7
Chasing the Tail2026.0727
SVRGSub-reward (SR)=false2026.0726
Base policyBackbone=Qwen2.5-3B-Instruct2026.0725.6