Our Single-rollout Asynchronous Optimization (SAO), is able to train stably for one thousand steps and consistently outperform GRPO and its variants on agentic coding and reasoning benchmarks,
such as SWE-Bench Verified, BeyondAIME, and IMOAnswerBench. https://t.co/oca1DvDvdB
SWE-bench — The standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
Why it matters
Zhipu's SAO training method trains stably for a thousand steps and beats GRPO variants on SWE-benchThe standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.Full definition →, BeyondAIME, and IMOAnswerBench, a concrete RL technique for agentic and reasoning modelA model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.Full definition → builders.
articleSERPO: Self-Evolving Rubric Policy Optimization for Open-Ended Test-Time Reinforcement LearningJianze Wang, Kunwang Zheng, Ying Liu, Yu Cao, Qilong Zhang, Jinlong Chen, Hua Yang, Qianglong Chen
articleRL Systems: Mind the Gap — Matching Trainer and Generator ThroughputKimbo Chen