<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Code-Reasoning on danilchenko.dev</title><link>https://www.danilchenko.dev/tags/code-reasoning/</link><description>Recent content in Code-Reasoning on danilchenko.dev</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Wed, 13 May 2026 08:24:46 +0000</lastBuildDate><atom:link href="https://www.danilchenko.dev/tags/code-reasoning/index.xml" rel="self" type="application/rss+xml"/><item><title>THINC: How a 4B Model Beat 235B Qwen3 by Reasoning in Code</title><link>https://www.danilchenko.dev/posts/thinc-code-reasoning/</link><pubDate>Wed, 13 May 2026 08:24:46 +0000</pubDate><guid>https://www.danilchenko.dev/posts/thinc-code-reasoning/</guid><description>THINC trains a 4B parameter model to reason entirely in code. It scored 78.1% on competition math, beating Qwen3-235B at 75.2%. Here&amp;#39;s how the method works.</description></item></channel></rss>