Qwen3-Coder 30B-A3B Instruct

Qwen3-Coder 30B-A3B Instruct is the smaller open MoE variant in the Qwen3-Coder family. It has 30.5 billion total parameters, activates only 3.3 billion per token, and supports non-thinking mode only. A purpose-built function-calling format connects the checkpoint with Qwen Code, Cline, and similar coding environments.

Native context reaches 262,144 tokens and can be extended to 1 million with YaRN. Compared with Qwen3-Coder 480B-A35B Instruct, this checkpoint activates less than one-tenth as many parameters per token.