Gemma 4 audio with MLX

Simon Willison published a working command-line recipe on April 12, 2026, that lets Mac users transcribe audio files locally using Google's Gemma 4 E2B model through Apple's MLX framework. This matters because it puts a capable, 10.28 GB multimodal AI model on consumer hardware without any cloud API calls, cloud costs, or data leaving your machine.

This story matters because it signals a shift in how AI agents are being adopted across the industry. For builders evaluating their AI stack, this is worth watching closely.

Get stories like this daily

Free briefing. Curated from 50+ sources. 5-minute read every morning.