Running DeepSeek v4.1 locally on M5 Max at 17 tokens/s, project open-sourced

Argonautlabs · hn · 2026-09-16

Argonautlabs shared a test of running DeepSeek v4.1 locally on an M5 Max, achieving about 17 tokens/s, and open-sourced the related project argodrive on GitHub. A useful data point for anyone tracking Mac local LLM inference performance.

Original post →

More from Infra

Infra channel →