danveloper/flash-moe: Pure C/Metal Inference Runs Qwen3.5-397B on MacBook Pro 48GB — 1.8K Stars, 393 HN Points

GitHub·medium signal

Built by a CVS Health VP of AI, flash-moe is a pure C/Metal inference engine that runs the 397B-parameter Qwen3.5 MoE model on a MacBook Pro with 48GB RAM. Created March 18, it reached 1.8K stars and 393 HN points by demonstrating that frontier-class MoE models can run on consumer hardware without Python dependencies. Directly validates the local inference economics thesis for builders who want to avoid API costs.