Qwen3-Coder 30B-A3B Instruct

Qwen3-Coder-30B-A3B-Instruct is a coding-focused large language model released by the Qwen team as part of the broader Qwen3-Coder family, which ships in multiple sizes. It is positioned as a streamlined variant that delivers strong results on agentic coding, agentic browser-use, and foundational coding tasks, making it well suited for developers building automated software workflows. The model is designed around a specially crafted function-call format and is compatible with widely used agentic platforms such as Qwen Code and Cline, giving it practical reach across agent-style integrations.

Under the hood, Qwen3-Coder-30B-A3B-Instruct is a causal language model trained with a combination of pretraining and post-training stages. It uses a Mixture-of-Experts architecture with 30.5 billion total parameters and 3.3 billion activated parameters spread across 48 layers, 128 experts, and grouped-query attention with 32 query heads and 4 key-value heads. It natively handles very long inputs of up to 256K tokens and can be extended to roughly 1M tokens through YaRN scaling, which is particularly valuable for repository-scale understanding and whole-codebase reasoning. Because it operates in a non-thinking mode without generating chain-of-thought blocks, the model is oriented toward direct, tool-driven coding assistance rather than open-ended reflective reasoning.