Asynchronous Single-Rollout RL for Agents
de4dee · reddit · 2026-07-14
The paper proposes Single-Rollout Asynchronous Optimization (SAO) to address the stability and off-policy issues of asynchronous RL in long-horizon agent tasks.
Main Approach
- Replaces the intra-group sampling common in GRPO with a single rollout to reduce off-policy deviation during asynchronous training
- Introduces a more practical value model training design
- Incorporates strict two-sided token-level clipping to improve optimization stability
Results
- Training stably runs for 1,000 steps and consistently outperforms GRPO and its variants on benchmarks like SWE-Bench Verified, BeyondAIME, and IMOAnswerBench
- Also performs better in simulated online learning scenarios, demonstrating adaptability to changing environments
- This method has already been applied to the agentic RL training pipeline for the open-source GLM-5.2
Related event: SAO Outperforms GRPO in Coding and Reasoning Benchmarks(3 posts)→
More from coding & agent
- DeepMind paper: LLMs can't sample randomly on their own, only convert given seeds — rohanpaul_ai · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27
- BELL: Open-Source Stack Combines Browserbase, Eve, iMessage and Stripe Link for Personal Agents — jeff_weinstein · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Two engineers rebuilt a FedEx supplier's billing core in 3.5 months vs an estimated 7-8 — alex_verem · 2026-10-27
- GPT-6.1 Terra scores 48 on AA Coding Agent Index at one-third the cost of GPT-5.6 Terra — ArtificialAnlys · 2026-10-27