# Researchers report direct weight edits reduced Qwen reasoning loops

_Published Monday, November 2, 2026 at 2:08 PM EST · AI · Latest · Tier 2 — Notable_

![Researchers report direct weight edits reduced Qwen reasoning loops — Primary](https://rikiyat.github.io/images/blog/weight-space-programming.png)

Researchers report that GPT-6 Astra improved Qwen3.5-0.8B's benchmark results by directly editing its numerical weights without gradient-based training. With 50 evaluation runs, MATH-500 rose from 7.9% to 48.1%, while instruction-following scores on IFEval rose from 21.0% to 62.3%.

The experimental method changed 131,072 weights, or 0.016%, to reduce repetitive reasoning that prevented answers from finishing. The researchers released experiment logs, code and evaluation records. They caution that removing loops exposed existing ability rather than adding mathematical skill; the method failed to establish an improvement on Gemma 3 270M, which rarely loops.

## Sources

- [rikiyat.github.io](https://rikiyat.github.io/blog/weight-space-programming/)

---
Canonical: https://techandbusiness.org/newswire/oxbGVGU45KtSRwtafwPk7T
Published: 2026-11-02T19:08:06.445Z
Story chronology: 2026-11-02T08:00:00.000Z
Retrieved: 2026-11-02T21:47:45.779Z
Publisher: Tech & Business (techandbusiness.org)