Qwen3.8-27B coding model ported to MLX for Mac

The RL-tuned Qwen3.8-27B-opencode-adapted model has been ported to Apple Silicon via MLX with 4-bit quantization, reaching up to 156 tok/s and 73 tok/s on M5 Max, enabling high-throughput local coding inference on Mac.

2026-11-19 ~ 2026-11-19 · 2 related posts