# Best way to run Qwen3.8 27B on M4 Pro 48gb? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/80b7608bbde4d26a

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-25 07:14 ingested 2026-10-26 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xevqqs/best_way_to_run_qwen38_27b_on_m4_pro_48gb/)

## 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
