The FASTEST Way to Run Qwen 3.6 27B on a Mac - MTPLX Setup + Hermes Agent

Qwen 3.6 27B runs 3x faster on a Mac with MTPLX — 66.3 tokens per second on a dense 27B model, no quality loss and no separate draft model. In this MTPLX setup guide I walk through installation on Mac Studio and MacBook Pro, benchmark the speed difference before and after, and then run it as a local server wired straight into Hermes Agent for a fully offline agent stack.

Video chapters

• 0:00 — The Fastest Way to Run Qwen 3.6 27B on a Mac
• 1:16 — MTPLX Setup Guide
• 4:45 — MTPLX Settings
• 7:37 — Setting Up MTPLX as a LLM Server
• 10:16 — Starting The Server
• 10:46 — Model Settings (Qwen 3.6 27B)
• 13:05 — Hermes Agent Model Setup

Watch the full video and find the original resources on YouTube.