← sources

Best way to run Qwen3.8 27B on M4 Pro 48gb?

source: reddit (reddit:LocalLLaMA)published 2026-10-25 07:14ingested 2026-10-26 02:40

retained text

I’m looking for a setup with the best balance of tps, ttft, context length and enough headroom for other tasks on the Mac.

I’m getting a lot of conflicting opinions with regards to MTP,  MLX, quantization etc.  Please don’t recommend options less than 10 tps. 

document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xevqqs/best_way_to_run_qwen38_27b_on_m4_pro_48gb