Qwen 3 Coder 480b

• 🧠 480B-parameter MoE with 35B active experts
• 📏 Massive 256K token context window
• 🔧 Native function calling and web search
• 📜 Apache 2.0 open-source license
• 🎯 Alibaba's flagship code-specialized model
• 🧩 FP8 quantization for efficient serving

Alibaba

Alibaba Group is a Chinese multinational technology company founded in 1999 and headquartered in Hangzhou, Zhejiang. Originally built around e-commerce and cloud computing, Alibaba has become one of the most prolific contributors to open-weight AI research,…

Qwen 3 Coder 480b is Alibaba's flagship code-specialized large language model, a 480-billion-parameter mixture-of-experts design that activates roughly 35B parameters per token for efficient inference at frontier scale. Served here in FP8 quantization under the permissive Apache 2.0 license, it pairs a 256K context window with native function calling and web search, making it well suited to agentic coding workflows, long-repository reasoning, and multi-file refactors that would overwhelm smaller context budgets.

Within Alibaba's Qwen lineup, the 480B Coder sits at the top of the coding-focused tier and is the default code model in the catalogue. It is later joined by a faster variant, Qwen 3 Coder 480B Turbo, which targets the same weights with lower-latency serving, while general-purpose reasoning is handled by siblings like Qwen 3 235B Thinking and the newer Qwen 3.5 397B. The Coder line remains a distinct, purpose-built branch rather than being superseded by those general models.

It's best used for autonomous coding agents, large-context code generation and review, tool-driven development tasks, and any workflow that benefits from an open-weights model with strong software engineering priors.