Qwen3-Coder 30B-A3B
Alibaba · Code · MoE 30B-A3B · 262k context · Released 14 May 2025
Commercial use permitted
Open weights
Runs on CPU
Apple Silicon
Qwen3-Coder 30B-A3B is purpose-built for agentic software engineering rather than chat. It runs in a non-thinking mode by design, activating 3B of its 30B parameters per token so it generates quickly, and its 256k context window (up to 1M with extrapolation) suits working across a whole repository. It is a practical local counterpart to the much larger Qwen3-Coder 480B.
Strengths
- Strong at agentic coding, tool use, and multi-file edits
- Very long 256k native context, useful for whole-repository work
- Fast generation for its quality, thanks to the mixture-of-experts design
- Apache 2.0, so no commercial-use conditions
Weaknesses
- Needs the memory of a 30B model despite only 3B being active
- Non-thinking by design, so it will not show step-by-step reasoning
- Qwen does not publish headline scores on the card, so benchmark on your own code
Hardware requirements
| Quantisation | Approx. VRAM | Notes |
|---|---|---|
| Q4_K_M | ~18GB | Fits a 24GB card with some room for context |
| Q8_0 | ~32GB | Near-lossless, needs 36GB or more |
| FP16 | ~61GB | Full precision, server or multi-GPU territory |
Also runs on CPU (slower). Optimised builds available for Apple Silicon.
Licence
Apache 2.0 — read the licence
Availability
- Official page
- Hugging Face
- ollama run qwen3-coder:30b
Recommended for
- Local agentic coding on a 24GB card
- Working across large codebases where long context helps
- A permissively licensed coding model for commercial work
Related guides
Glossary
Catalogue entry last verified 30 July 2026. Specifications change; verify anything you are about to spend money on.