Qwen3-Coder has been trained using a combination of pre-training and post-training methods. During pre-training, the model was scaled along multiple dimensions to strengthen its core capabilities, including scaling tokens, context, and synthetic data. The model was trained on 7.5T tokens with a 70% code ratio, excelling in coding while preserving general and math abilities. Additionally, the model natively supports 256K context and can be extended up to 1M with YaRN, optimized for repo-scale and dynamic data.
Qwen3-Coder can be used with various tools, including Qwen Code, a research-purpose CLI tool adapted from Gemini CLI. Qwen Code has been enhanced with customized prompts and function calling protocols to fully unleash the capabilities of Qwen3-Coder on agentic coding tasks. Additionally, Qwen3-Coder can be used with Claude Code, a popular coding tool. The model can also be accessed through the Alibaba Cloud Model Studio API, allowing developers to integrate it into their applications.
