From the model card
Fields below are copied from the tags and counters on the HuggingFace repository Qwen/Qwen3-Coder-30B-A3B-Instruct at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- Qwen
- Pipeline tag
- text-generation
- Library
- Transformers
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Papers cited
- arXiv:2505.09388
- Downloads (HF counter at last fetch)
- 698,536
- Likes (HF counter at last fetch)
- 1,226
- Model card
- https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct
Use cases
- Agentic code generation for multi-file software projects
- Code review and bug detection in CI pipelines
- Repository-level refactoring with context over large codebases
- Instruction-following for complex programming tasks with tool calling
Pros
- MoE efficiency: 3B active parameters give cost near a 3B model
- 30B total capacity stores more code knowledge than dense 7–14B models
- Apache-2.0 licensed
- Designed explicitly for agentic coding, not just completion
Cons
- All 30B parameters must reside in memory despite 3B being active
- Expert routing can be inconsistent across different programming languages
- Limited third-party fine-tuning guides compared to Llama-based code models
- Performance on low-resource languages (Rust, Zig) less evaluated
Tags
transformerssafetensorsqwen3_moetext-generationconversationalarxiv:2505.09388license:apache-2.0endpoints_compatibleregion:usdeploy:sagemakerdeploy:azure