Scaling Coding-Agent RL to 32x H100s. 160% Improvement on Stanford's TBenchgithub.com/Danau5tin2 pointsDanau5tin8 months ago