Qwen: Qwen3 Coder 480B A35B

Qwen3 Coder 480B A35B is a sparse mixture-of-experts code model designed to handle the kinds of tasks that arise inside a real software project, not just isolated snippets. Its 480 billion total parameters are distributed across 160 experts, but only 35 billion activate per forward pass, letting the model keep deep coding knowledge available without firing everything for every token. This architectural choice makes it practical to run at full strength while maintaining a large context window that can absorb an entire repository. The model is built for agentic workflows—reading files, running tools, debugging failures, and iterating across a codebase—which means it is tuned for multi-step software engineering rather than single-prompt completions.

The model is Alibaba's flagship open-source coding release and reaches state-of-the-art performance among open-weight models on established benchmarks like SWE-Bench Verified. Its training balances deep coding specialization with general reasoning and mathematics, so it can handle everything from system-level architecture decisions down to fine-grained debugging. Available through multiple providers on AI Gateway, it plugs into developer toolchains such as Claude Code, Cline, and Qwen Code for coding copilots and repository-scale workflows. The combination of a broad context window, selective expert activation, and open weights positions it as a practical foundation for teams that want powerful autonomous programming without locking into a single vendor ecosystem.