# Self-HPO preprint reports higher coding success by optimizing model and tools together

_Published Monday, October 19, 2026 at 12:40 PM EDT · AI, Science · Latest · Tier 2 — Notable_

Researchers report in an arXiv preprint that Self-HPO, a framework that alternates improvements to a coding model and the software that manages its interaction with its environment, raised Qwen3.5-35B-A3B's average resolve rate from 57.8% to 64.8% across three coding benchmarks.

The model analyzes its failures and proposes changes to that surrounding software, called a harness. Researchers then update the model through reinforcement learning under the selected harness, using difficult training tasks to guide the next round of changes. The authors report gains beyond the training distribution and stronger results than optimizing either component alone. These are benchmark findings, rather than evidence of production deployment.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2610.22791)

---
Canonical: https://techandbusiness.org/newswire/-DQtTGbzN3jb9lOn3bRHPx
Published: 2026-10-19T16:40:58.692Z
Story chronology: 2026-10-19T04:00:00.000Z
Retrieved: 2026-10-19T19:21:53.087Z
Publisher: Tech & Business (techandbusiness.org)