<?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>Mlx on Local First AI</title><link>https://localfirstai.eu/tags/mlx/</link><description>Recent content in Mlx on Local First AI</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Tue, 09 Jun 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://localfirstai.eu/tags/mlx/index.xml" rel="self" type="application/rss+xml"/><item><title>Same Hardware. Different Runtime. Same Result.</title><link>https://localfirstai.eu/posts/2026-06-09-mlx-vs-ollama-runtime/</link><pubDate>Tue, 09 Jun 2026 00:00:00 +0000</pubDate><guid>https://localfirstai.eu/posts/2026-06-09-mlx-vs-ollama-runtime/</guid><description>MLX and Ollama both run gemma4:26b on Mac Mini M4 Pro. Neither cliffs through 40K tokens. The Flash Attention cliff from Exp 007 was an Ollama implementation artefact, not a hardware property — now confirmed by an independent runtime.</description></item></channel></rss>