User tunes Qwen 3.8 27b on M5 Max, criticizes Opus instruction following

julianharris · x · 2026-08-16

A user is optimizing settings for Qwen 3.8 27b on an M5 Max (128GB), balancing factors like MTP, KV optimizations, resolution, and a 128k context window for coding. They also noted thermal throttling on the 14" chassis. Additionally, the user criticized the Opus model's instruction following capabilities, stating it remains mediocre and requires constant supervision compared to previous versions.

Original post →

More from coding & agent

coding & agent channel →