aiqbench analysis
This research demonstrates that Evolution Strategies (ES) for LLM post-training can effectively utilize asynchronous rollout generation without significant performance loss, addressing a key bottleneck in long-horizon agentic tasks. By tolerating moderate policy staleness, this approach offers a scalable alternative to synchronous methods and potentially reduces training costs compared to gradient-based techniques like GRPO. The findings suggest that ES-based optimization is viable for complex, multi-turn coding benchmarks where rollout times vary widely.
Asynchronous Is Nearly Free for Evolution Strategies on Long-Horizon Agentic Tasks
arXiv:2610.04196v1 Announce Type: new Abstract: LLM-based long-horizon agentic post-training is often bottlenecked by rollout generation: trajectories span many interaction turns, completion times vary substantially, and synchronous update barriers leave faster workers waiting for stragglers. Asynchronous reinforcement learning which has been adopted in LLM post-training addresses this inefficiency by consuming trajectories as they arrive, but introduces policy lag and off-policy optimization. Evolution strategies (ES) offer a backpropagation-free alternative for LLM post-training, yet it relies on a larger number of rollouts and existing practices have remained largely synchronous. In this short-form paper, we introduce bounded-staleness asynchronous ES and demonstrate it on Endless Terminals benchmark using Qwen2.5-7B-Instruct. Across three evaluation seeds, natural Async-1 matches synchronous ES, achieving 25.9\% versus 25.4\% held-out success. Controlled schedules that delay 10\% of each update cohort by four or eight policy updates reduce success by only 1.6 and 3.1 percentage points, respectively, without explicit off-policy correction. GRPO performs better overall, reaching 29.0\% held-out success, but importantly our results show that ES tolerates moderate policy staleness with limited degradation, opening possibilities for future improvement of ES-based post-training with asynchronous algorithms. To the best of our knowledge, we are the first to demonstrate the effectiveness of sync and async ES on a multi-turn terminal style agentic coding task.
Abstract:LLM-based long-horizon agentic post-training is often bottlenecked by rollout generation: trajectories span many interaction turns, completion times vary substantially, and synchronous update barriers leave faster workers waiting for stragglers. Asynchronous reinforcement learning which has been adopted in LLM post-training addresses this inefficiency by consuming trajectories as they arrive, but introduces policy lag and off-policy optimization. Evolution strategies (ES) offer a backpropagation-free alternative for LLM post-training, yet it relies on a larger number of rollouts and existing practices have remained largely synchronous. In this short-form paper, we introduce bounded-staleness asynchronous ES and demonstrate it on Endless Terminals benchmark using Qwen2.5-7B-Instruct. Across three evaluation seeds, natural Async-1 matches synchronous ES, achieving 25.9\% versus 25.4\% held-out success. Controlled schedules that delay 10\% of each update cohort by four or eight policy updates reduce success by only 1.6 and 3.1 percentage points, respectively, without explicit off-policy correction. GRPO performs better overall, reaching 29.0\% held-out success, but importantly our results show that ES tolerates moderate policy staleness with limited degradation, opening possibilities for future improvement of ES-based post-training with asynchronous algorithms. To the best of our knowledge, we are the first to demonstrate the effectiveness of sync and async ES on a multi-turn terminal style agentic coding task.
Artificial Intelligence (cs.AI)
arXiv:2610.04196 [cs.AI]
(or arXiv:2610.04196v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.04196 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 297 words
Related coverage
[AINews] not much happened today
Latent Space · Sep 18, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
[AINews] DeepSeek v4.1-Flash: 763B-P8B-D16B novel causal Encoder–Decoder architecture with vision marks the Return of the Whale
Latent Space · Sep 12, 2026
How Sonos rebooted itself
The Verge · Sep 3, 2026