Qwen3.8-Flash-Next-MLX-Serve-mixed-4 · mlx-serve · Mac Studio/MBP M4 Max 128GB
English | 简体中文 | 日本語 · ← All recipes
Source level: 📣 Claimed · Date: 2026-08-27 · Numbers come from the source. This repo did not measure them again.
Hardware
- Mac Studio/MBP M4 Max 128GB (Mac)
- Reference cost: Not published. See the source.
Model
- Qwen3.8-Flash-Next-MLX-Serve-mixed-4 ·
mixed 4/8-bit (≈75 GB resident) - Download: huggingface.co/ddalcu/Qwen3.8-Flash-Next-MLX-Serve-mixed-4-8bit
Engine
- mlx-serve
- Note: mlx-serve 26.8.11
Launch command
mlx-serve --model ddalcu/Qwen3.8-Flash-Next-MLX-Serve-mixed-4-8bit --serve
Key parameters
- Context: 268,288
- Concurrency: serial single request
- Tags:
4bit,8bit,MoE,long-context
Expected speed
- 60.0 tok/s decode; 730.0 tok/s prefill (📣 Claimed)
Caveats
- Not a first-hand measurement (claimed or relayed). Your results can differ.
- Speed changes with drivers, engine version, context length, and prompt length.
Source & attribution
- https://huggingface.co/ddalcu/Qwen3.8-Flash-Next-MLX-Serve-mixed-4-8bit
- Credit to the original author. This page only summarizes and links. Data under CC BY 4.0.
Source file: recipes/hf-ddalcu-qwen3-8-flash-next-mlx-serve-mixed-4-8bit-1.md