<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Openspec on QueryMT</title><link>https://query.mt/blog/tags/openspec/</link><description>Recent content in Openspec on QueryMT</description><generator>Hugo</generator><language>en</language><lastBuildDate>Wed, 11 Nov 2026 18:00:00 +0100</lastBuildDate><atom:link href="https://query.mt/blog/tags/openspec/index.xml" rel="self" type="application/rss+xml"/><item><title>Does AREX-2 actually self-improve?</title><link>https://query.mt/blog/posts/arex2-self-improving-tsp/</link><pubDate>Wed, 11 Nov 2026 18:00:00 +0100</pubDate><guid>https://query.mt/blog/posts/arex2-self-improving-tsp/</guid><description>AREX-2 against its own base model, Qwen3.8-27B, and against GLM-5.3-flash, on one optimisation task with an external scorer. Nine runs on two local GPUs, with charts, the scoring harness, and the reasoning traces where AREX-2 corrects itself.</description></item></channel></rss>