FastRouter
Alibaba has released Qwen3-Coder-Next, an open-source 80B-parameter coding model that activates just 3B parameters per query, scoring 70.6% on SWE-Bench.
Model details
Qwen3 Coder is positioned as a coding-focused large language model designed to power agent-style developer workflows and tool-augmented code generation. It is part of the broader Qwen family of open-weight models, making it accessible for self-hosting and integration into custom coding assistants. The model is designed to interact through tool calling, giving it the ability to chain operations, invoke external utilities, and participate in multi-step engineering tasks rather than producing isolated code snippets.
The broader Qwen3-Coder line explored several architectural directions, including sparse mixture-of-experts configurations where small subsets of parameters are activated per query to balance capability and compute cost. The lineage includes a large 480B-class instruct variant made available on accelerated inference hardware, and a more compact 80B-parameter variant that activates only 3B parameters per forward pass while still reporting competitive coding benchmark results. These follow-on releases suggest an emphasis on giving developers flexible deployment options, from full-scale hosted inference to efficient local or edge-oriented coding agents.
Transparent token rates
Rates are shown per one million tokens. Combined means one million input plus one million output tokens.
FastRouter
Alibaba has released Qwen3-Coder-Next, an open-source 80B-parameter coding model that activates just 3B parameters per query, scoring 70.6% on SWE-Bench.
FastRouter
Alibaba's Qwen3 Coder 480B Instruct model is now available on Cerebras.