Best rig for Qwen3-Coder-Next
Qwen3-Coder-Next is an 80B-parameter MoE with ~3B active. The activation pattern means it punches above its weight on Apple Silicon and sane consumer GPUs. Here's the data we have.
We have no measured Qwen3-Coder-Next run yet, so this guide can't recommend hardware. The closest measured model with the same 80B-A3B shape is Qwen3-Next-80B-A3B 4-bit, at 80.3 tok/s on a 96GB M3 Ultra (MLX, r_1pl79r50ofy). That is a different model, shown only as a reference point.
No data submitted for this task yet.
Run the suite to be the first benchmark for this guide:
$ pipx install llm-speed && llm-speed bench
Qwen3-Coder-Next's 80B-A3B mixture-of-experts means raw VRAM matters less than for a dense model of comparable footprint, but you still need ~40-50 GB of memory for a 4-bit quant. That puts it in reach of an M-series Ultra, a pair of 24 GB consumer GPUs, or a single MI300X / H100. Below is every config we have a real submitted run for.
Check it for your setup
- Predict your exact setup: estimate whether Qwen3-Coder-Next fits on M3 Ultra at your quant and context, next to measured speeds where we have them.
- Request this pair: ask for a measured Qwen3-Coder-Next run on M3 Ultra. Requests are ranked by votes.
- Full run table (0 runs): every Qwen3-Coder-Next result we have, with links to each source run.
See also: All hardware · All models · Methodology