Alibaba has released Qwen3-Max-Preview, a model with over a trillion parameters.

Recently, Alibaba unveiled its largest language model to date, Qwen3-Max-Preview. With over a trillion parameters, the model sets a new benchmark in AI technology. Qwen3-Max-Preview is now accessible to users through Qwen Chat and the Alibaba Cloud API. According to Alibaba, the model has outperformed its previous top-tier model, Qwen3-235B-A22B-2507, in both internal benchmarks and early user feedback.

Key Features and Performance

Qwen3-Max-Preview shows significant improvements in several areas, including knowledge comprehension, conversational skills, task handling, and instruction execution. The model has also effectively reduced “knowledge hallucinations,” meaning it provides more accurate and reliable information.

The model can process up to 258,048 input tokens and generate up to 32,768 output tokens. This high capacity makes Qwen3-Max-Preview an excellent performer in complex conversations and tasks. For users who need it, the pricing is set at $2,151 per million input tokens and $8,602 per million output tokens. It’s important to note that Qwen3-Max-Preview currently does not support image processing.

Future Outlook

With these powerful features, Alibaba aims to further advance the field of natural language processing, offering more intelligent AI solutions to users. This innovative language model is expected to bring more possibilities for businesses and developers in their applications, helping them succeed in their digital transformation journey.