List your company

Post-RL Evaluation on BiGGen Bench

63.7AccuracyMRRG
57.56459.15760.7562.343Jul 2, 2026

Evaluation results

MethodLinks
MRRG2026.0763.7
SVRGSub-reward (SR)=true2026.0762
Chasing the Tail2026.0761.1
SVRGSub-reward (SR)=false2026.0760
Base policyBackbone=Qwen2.5-3B-Instruct2026.0757.8