Qwen3-Coder

Qwen3-Coder is the code iteration of the Qwen3 large language model series developed by the Qwen team. This model is particularly designed for agentic coding tasks and comes in various sizes, offering performance in both coding and agentive functions. The Qwen3-Coder-Next model has been specially trained on large-scale executable task synthesis, environment interaction and reinforcement learning. This advanced training has provided it with robust coding and agentive functions while keeping inference costs significantly lower. A significant feature of Qwen3-Coder is its efficiency-performance trade-off, making it highly effective in foundational coding tasks and agentic browser use. It can support most platforms including Qwen Code, CLINE, and Claude Code, and features a function call format designed specifically for it. Qwen3-Coder can handle long-context capabilites with inherent support for 256K tokens, extendable up to 1M tokens using Yarn. This allows it to optimize for repository-scale understanding. The model can understand and generate in a long context length of 256K tokens and supports 358 coding languages.