Qwen3 235B A22B Instruct

Qwen3 235B A22B Instruct is the flagship model in Alibaba's Qwen3 family, using a Mixture-of-Experts architecture with 235 billion total parameters and 22 billion active per token. This MoE design delivers high-quality outputs with efficient inference, activating only a fraction of parameters for each input.

The model excels across reasoning, coding, mathematics, multilingual understanding, and instruction following. It supports an extended context window for processing long documents and complex multi-turn conversations, with strong tool-use capabilities for agentic workflows.

Qwen3 235B represents Alibaba's most capable open model, competing with frontier models on major benchmarks while offering efficient inference through its MoE architecture.