AI News (2026/1/27): Qwen3-Max-Thinking, the Most Powerful Model, Officially Released
Executive Summary:
Alibaba Cloud has officially released its largest and most capable inference model, Qwen3-Max-Thinking. The model has over 1 trillion parameters and 36T Tokens of pre-training data, setting new records in multiple international benchmark tests. Qwen3-Max-Thinking innovatively adopts a test-time expansion mechanism, enhancing inference performance while being more economical.
Qwen3-Max-Thinking, the Most Powerful Model, Officially Released
News Details
Alibaba Cloud officially released its latest inference model, Qwen3-Max-Thinking, on Tuesday, January 27. This model is the largest and most capable AI model to date from Alibaba Cloud, with over 1 trillion parameters and 36T Tokens of pre-training data. Qwen3-Max-Thinking has set new records in multiple international benchmark tests, demonstrating its exceptional performance in inference tasks.
Key Points
Parameter Count: Qwen3-Max-Thinking has over 1 trillion parameters, significantly surpassing other large-scale models currently on the market. This increase in parameter count enhances the model's accuracy and robustness when handling complex tasks.
Pre-training Data: The model's pre-training data amounts to 36T Tokens, covering a wide range of languages and domain knowledge. The rich pre-training data helps the model perform well in various application scenarios, especially in low-resource and long-tail tasks.
Test-time Expansion Mechanism: Qwen3-Max-Thinking innovatively adopts a test-time expansion mechanism (Test-time Expansion), dynamically adjusting the model size during inference to balance performance and cost. This mechanism ensures that the model maintains high performance while being more cost-effective, making it suitable for large-scale deployment.
AI-ALL In-depth Analysis
The release of Qwen3-Max-Thinking marks another significant breakthrough for Alibaba Cloud in the field of large-scale language models. With over 1 trillion parameters and 36T Tokens of pre-training data, the model not only enhances inference capabilities but also achieves notable results in multiple international benchmark tests. The introduction of the test-time expansion mechanism provides a new solution for the practical application of large-scale models, making it possible to balance high performance with low cost. This is an important development for AI developers and enterprises, especially in resource-constrained environments, allowing for more flexible selection and deployment of appropriate model sizes. Additionally, the release of Qwen3-Max-Thinking further promotes the popularization and application of AI technology, laying a solid foundation for the future development of the AI ecosystem.