Comment by gruez

>Opus 4.6 Max self-hosted at 30 tok/s on a 5k Macbook in Aug 2026. The LLM timelines are crazy.

How much memory does this translate to and what quantization (if any) were applied?