Qwen is the series of large language models and multimodal models made by the Qwen Team of Alibaba Group. A large language model (LLM) is a neural network trained on very large amounts of text to predict the next token, so that it can answer questions, write text and write code. The Qwen models handle text generation, tool use, image and audio understanding, and more than 100 languages.
Qwen3-Coder is the code version of Qwen3. Its largest variant, Qwen3-Coder-480B-A35B-Instruct, is a Mixture-of-Experts (MoE) model: of its 480 billion parameters, only 35 billion are active for each token, because each token goes to 8 of its 160 expert sub-networks. It reads 256K tokens of context natively, extendable to 1M, and it works inside coding agents such as Qwen Code, Cline and Claude Code through a special function call format.
The weights are open: they are published on Hugging Face and ModelScope under the Apache 2.0 license and run locally with tools such as vLLM, SGLang, llama.cpp and Ollama. Alibaba Cloud Model Studio also sells the models through an API, billed per million tokens.