On November 16, Alibaba's Tongyi Qianwen team officially released Qwen4-Max, moving the model from a limited preview to a general release. Developers can now call it through Token Plan or open APIs on the Qianwen AI platform. The model has 2.4 trillion total parameters in a sparse mixture-of-experts architecture, activates 71 billion parameters per token, supports native vision understanding and offers a one-million-token context window.
Rather than simply cutting prices, Alibaba focused on reducing token consumption. In the Terminal-Bench 4.0 evaluation, average token usage per attempt fell from 289,000 to 151,000, while the score rose from 37.9 to 48.5. In the Work360 benchmark, the average cost of completing a single task dropped from 26.58 yuan to 11.16 yuan, a reduction of 58 percent. Cache-hit pricing is as low as 0.5 yuan per million tokens.
The model has been integrated into Qianwen Office, Qoder, the Qianwen app and Qwen Studio, and is available on Alibaba Cloud's Bailian platform in both mainland China and global versions, compatible with OpenAI and Anthropic API protocols. Smaller Qwen4 models are expected to be open-sourced gradually.
Analysts said the launch signals a shift in large-model competition from parameter counts and benchmark scores toward the computing and token cost of completing a real-world task.





