MTPLX V2 Boosts Local Mac LLM Inference to 80+ TPS

Optimized for Apple Silicon, MTPLX V2 introduces a Turbo mode with custom quantized matrix multiplication kernels, achieving over 80 tokens per second running Qwen 3.6 27B on MacBook Pro M5 Max.

2026-07-08 ~ 2026-07-09 · 2 related posts

• MTPLX V2 Released: Runs Qwen 27B on MacBook M5 Max at 72+ Tokens/sec — awnihannun · 2026-07-08
• MTPLX V2 Released: MLX Inference Hits 82 TPS on Mac — YoussofAl · 2026-07-09