List your company

Held-out Skill Evaluation on Multi-family n=225 (holdout)

43.5556Solve RateGRPO from SFT init
10.27555618.91555627.55555636.195556May 27, 2026

Evaluation results

MethodLinks
GRPO from SFT initbatches=40, initialization=SFT init2026.0543.5556
GRPO from base initbatches=40, initialization=base init2026.0527.5556
Base Qwen3-8B2026.0511.5556