Alibaba launches Qwen4-Max with closed weights
The flagship AI model is available through an API and Alibaba apps, while smaller Qwen4 models are due to receive open weights.

Alibaba introduced its flagship Qwen4-Max AI model on November 16, making it available through hosted services rather than allowing users to download and modify it. The company said it would provide open weights for smaller Qwen4 models but gave no timetable.
The model handles text, images and video and accepts prompts containing up to one million tokens. Its mixture-of-experts architecture activates only a portion of its 2.4 trillion parameters for each request.
Developers can access it through the QwenCloud API, where each million input tokens costs $1.50 and each million output tokens costs $4.50. Alibaba also offers the model in Qwen Studio, QwenWork and Qoder. A restricted preview began on November 11.
Alibaba's benchmark results showed differing performance across coding tests. Its published charts put Qwen4-Max at 75.7 on DeepSWE v1.1, compared with 74.2 for Claude Opus 5.5. On Terminal-Bench 4.0, Qwen4-Max recorded 48.5, below Opus 5.5's 59.6 and Kimi K3.1's 67.2. TNW said it had not independently verified Alibaba's scores.
The company said Qwen4-Max used fewer tokens than Qwen3.8-Max to complete work. On Alibaba's internal Work360 evaluation, the mean cost per task declined from $4.43 to $1.67.
Alibaba said its training retained material supporting defensive cybersecurity work, including detecting and repairing software vulnerabilities, while excluding material intended to improve attack capabilities. It also said it added training intended to reduce chemical, biological, radiological and nuclear risks.
In a separate claim about the model's capabilities, Alibaba said Qwen4-Max spent 48 hours working independently on a chip-coolant material design that was subsequently produced and tested in a laboratory.




