Overview

• Alibaba's Qwen 3 series includes eight AI models, with the flagship 235-billion-parameter model outperforming DeepSeek R1 and OpenAI o1 on major benchmarks.
• The models introduce hybrid reasoning, allowing users to toggle between fast responses and in-depth problem-solving for complex tasks.
• Qwen 3 supports 119 languages and was trained on a dataset of 36 trillion tokens, doubling the size of its predecessor Qwen 2.5.
• Available under an open Apache 2.0 license, the models can be accessed via platforms like Hugging Face, GitHub, and Alibaba Cloud.
• This release highlights China's growing competitiveness in AI, driven by domestic innovation amid U.S. chip export restrictions.