Alibaba Releases Qwen3 Model Family Including 235B MoE Flagship

Alibaba's Qwen team open-sourced the Qwen3 family, ranging from Qwen3-0.6B to Qwen3-235B-A22B (a mixture-of-experts model activating 22B parameters per forward pass). The family supports 119 languages, features a hybrid thinking/non-thinking mode switchable via a chat tag, and is released under Apache 2.0. Benchmark results show Qwen3-235B-A22B matching or exceeding GPT-4.1 and Claude 3.7 Sonnet on several coding and math tasks. Alibaba Cloud captured more than 50% of global open-source model downloads as of March, with Qwen reaching nearly 1 billion cumulative downloads.