← Every modelAlibaba (Qwen) · 235B total, 22B active (mixture of experts)

Qwen3 235B-A22B

The largest Qwen3. It needs about 145 GB to load, so it is a model for multi-GPU rigs and the biggest Macs, but with 22B active parameters it runs at a usable speed once it fits.

ollama run qwen3:235b

Needs about 144 GB of graphics memory, or unified memory on a Mac. First time? Install Ollama, then paste that into a terminal.

Download142 GB
Memory to runAbout 144 GB
Read per token13.3 GB (mixture of experts)
LicenseApache 2.0
Context32K tokens, 128K with YaRN
ReleasedApril 2025
Good for
  • Hard coding and reasoning on a multi-GPU rig
  • Thinking mode on long problems
  • A step up when gpt-oss 120B isn't enough
Not the best pick for
  • Anything under about 150 GB of memory
  • Image input: it is text only
2 GPUs and Macs

What runs Qwen3 235B-A22B

Estimated speed on each card at the smallest memory size that fits it, from memory bandwidth. As a mixture-of-experts model it reads only part of its weights per token, so it runs much faster than its size suggests.

Usable 2 cards

Cheapest build from our list that runs it well: Dell DSS 8440 + 10× Tesla P100, from $5,100 used, at about 19 tok/s. See the builds.

Compare

Other models