Self-evolving coding agents through recursive optimization
In brief
- Self-HPO recursively optimizes both policy and environment for coding agents.
- Boosts Qwen3.5-35B-A3B's resolution rate from 57.8% to 64.8% across three benchmarks.
- Gains generalize to out-of-distribution tasks and exceed the sum of individual improvements.
A new framework, Self-HPO, enables LLM-based coding agents to improve by jointly optimizing their policy and environment. Results show significant gains across three benchmarks, with generalization to unseen tasks.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error