Alibaba has launched Qwen3-Coder, an open-source artificial intelligence model designed for high-performance software development. The model features a 480-billion-parameter Mixture-of-Experts (MoE) architecture with 35 billion active parameters. It natively supports a 256,000-token context length and can extrapolate to handle up to 1 million tokens, enabling extensive code generation and agentic tasks. Benchmark results indicate that Qwen3-Coder delivers performance comparable to Claude Sonnet 4 and surpasses other open-source models such as Kimi K2, DeepSeek, and Gemini 2.5 Pro on coding benchmarks like SWE-Bench Verified, achieving scores around 69.6% to 70%. The model is released under the Apache 2.0 license and is available on the Hugging Face Hub. Alongside the model, Alibaba has open-sourced Qwen Code, a command-line interface tool adapted from Gemini CLI, to facilitate agentic programming. Pricing for token usage starts at $1 per million tokens for input and ranges up to $60 per million tokens for output. Following this release, Alibaba also introduced Qwen3-235B-A22B-Thinking-2507, its most advanced reasoning model to date, with 235 billion parameters and 22 billion active parameters. This reasoning model supports a 256,000-token context and features improved performance in logical reasoning, math, science, coding, instruction following, tool use, and alignment. It has outperformed other open-source reasoning models such as Gemini 2.5 Pro and OpenAI's open models on benchmarks including AIME25 and GPQA. Both models reflect Alibaba's ongoing efforts to advance AI capabilities in coding and complex reasoning tasks.