Qwen3-Coder

Qwen3-Coder is Alibaba's open-source code model family, with Qwen3-Coder-480B-A35B-Instruct as its flagship variant. The 480 billion parameter mixture-of-experts architecture activates just 35 billion parameters per token, balancing frontier-class capability with practical inference costs. It natively supports a 256K context window and extrapolates to 1M tokens, making it one of the strongest open options for large codebase reasoning. Released under the Apache 2.0 license, the model posts competitive results against closed-source SOTA on agentic coding, browser interaction, and tool use benchmarks. Weights are on Hugging Face, inference is supported on Together AI, DashScope, and most major gateways. For teams that need a self-hostable, commercially usable coding model without vendor lock-in, this is the current open leader.