Overview

• Alibaba unveiled Qwen 3, a family of eight AI models, including two mixture-of-experts (MoE) models and six dense models ranging from 0.6 billion to 235 billion parameters.
• The Qwen 3 models support 119 languages, were trained on 36 trillion tokens, and feature hybrid reasoning capabilities with configurable thinking modes.
• Internal benchmarks show the flagship Qwen-3-235B-A22B model outperforming OpenAI’s o3-mini and Google’s Gemini 2.5 Pro in coding and math tasks.
• All Qwen 3 models are open source and available—or soon to be—on Hugging Face and GitHub, enhancing accessibility for developers worldwide.
• The series emphasizes cost efficiency, with Alibaba claiming significant reductions in deployment costs compared to other major AI models.