Qwen Code is a series of large language models (LLMs) optimized for software engineering tasks including code generation, reasoning, and debugging.

• Generates syntactically correct code across 92 programming languages, including Python, C++, Java, Go, and TypeScript.
• Processes long-context sequences up to 128,000 tokens to enable repository-level understanding and multi-file editing.
• Outperforms comparable models on industry benchmarks such as HumanEval, MBPP, and LiveCodeBench.
• Offers model scales from 0.5B to 32B parameters, supporting both local edge deployment and high-throughput server environments.

The models are built on a decoder-only transformer architecture and are compatible with frameworks like Hugging Face Transformers, vLLM, and Ollama under the Apache 2.0 license. Qwen Code is ideal for building AI-powered IDE extensions, automating technical documentation, and performing legacy code migration.