Alibaba Unveils Qwen3 AI Models with Hybrid Reasoning

Alibaba has launched Qwen3, a new series of AI models that feature hybrid reasoning capabilities, . The Qwen3 series includes eight models, ranging from 0.6 billion to 235 billion parameters, and is designed to compete with leading models from OpenAI and Google.

The Qwen3 models are available under an open-source license and can be accessed on platforms like Hugging Face and GitHub. They offer a mixture of experts (MoE) architecture, which allows for more efficient computation by breaking tasks into subtasks handled by specialized models. This design enables users to toggle between fast, general-purpose responses and more complex, compute-intensive reasoning steps.

Trained on a dataset of nearly 36 trillion tokens, the Qwen3 models support 119 languages, significantly expanding their potential applications globally. The largest model, Qwen-3-235B-A22B, has shown competitive performance in benchmark evaluations, surpassing models like OpenAI's o3-mini and Google's Gemini 2.5 Pro in certain tests.

The release of Qwen3 underscores the intensifying competition in China's AI sector, with Alibaba positioning itself as a leading provider of open-source AI models. The models are available for deployment across various platforms, including cloud providers like Fireworks AI and Hyperbolic.