Alibaba, the Chinese tech giant, has launched Qwen3, the latest generation in its open-sourced large language model family. Marking the company's first foray into hybrid reasoning, Qwen3 aims to fuel the next wave of AI-powered applications worldwide.
The Qwen3 series includes six dense models and two Mixture-of-Experts (MoE) models, giving developers the power to build advanced solutions across mobile devices, smart glasses, autonomous vehicles, robotics and more. These models are free to download and available globally.
What sets Qwen3 apart is its hybrid reasoning capability. It can switch to a "thinking mode" for complex, multi-step tasksâlike math, coding, and logical deductionâthen flip back to a "non-thinking mode" for lightning-fast, general-purpose responses.
Since the debut of its Qwen model family, Alibaba has seen over 300 million downloads worldwide. The launch of Qwen3 comes amid booming demand for generative AI in the Chinese mainland, where data shows nearly 200 models registered and launched by 2024, serving over 600 million users.
For global developers and innovators, the arrival of Qwen3 offers an open-source toolkit to push the boundaries of AIâwithout any download fees holding them back.
Reference(s):
Alibaba introduces free-to-download AI model with hybrid reasoning
cgtn.com
