Alibaba launches Qwen4-Max with 43 on Intelligence Index
Alibaba says Qwen4-Max cuts average cost per task to $3.27 and prices its API 25% below the previous model, according to aitimes.com.
Alibaba officially launched the Qwen4-Max AI model on November 16, publishing its architecture, benchmark results and API pricing on its official blog, according to aitimes.com. The release follows a preview service the company opened on November 11 without disclosing the model's structure or performance.
Qwen4-Max keeps the same 2.4 trillion total parameter scale as Qwen3.8-Max but reduces active parameters by 25%, from 95 billion to 71 billion. The model uses a mixture-of-experts architecture with those 71 billion active parameters plus 328 billion separate N-gram embedding parameters, and the Qwen4 architecture introduces Gated DeltaNet, Qwen Sparse Attention and Gated Residual.
On the Artificial Analysis Intelligence Index, Qwen4-Max scores 43, 9 points above Qwen3.8-Max's 34. Average cost per task falls from $10.02 to $3.27, a drop of about 67%. Qwen4-Max scores 75.7 on DeepSWE 1.1, ahead of Anthropic's Claude Opus 5.5 at 74.2, but trails Kimi K3.1 Pro on Terminal-Bench 4.0 with 48.5% against 67.2%.
On Alibaba's in-house Work360 benchmark, Qwen4-Max scores 52.69, up from 47.74 for the previous model. Token usage drops 24-35% depending on reasoning effort and about 48-49% on Terminal-Bench 4.0, while average Work360 cost per task falls from $4.43 to $1.67, about 62% lower.
API prices are 25% lower at $1.50 per million input tokens and $4.50 per million output tokens, versus $2 and $6 before. Cached input costs $0.075 per million tokens, down from $0.25. The model supports up to a 1 million token context and accepts text, image and video input, and is available as a pay-as-you-go API on Qwen Cloud, integrated into Qwen Work, Qwen Studio and Coder.
Alibaba did not release the model weights but said smaller Qwen4 series models will be released as open weights later, without giving sizes or dates.
Quick answers
How much does Qwen4-Max cost via API?
Alibaba priced it at $1.50 per million input tokens and $4.50 per million output tokens, 25% below the previous model. Cached input costs $0.075 per million tokens.
Where can Qwen4-Max be used?
The full model is available as a pay-as-you-go API on Qwen Cloud and is integrated into Qwen Work, Qwen Studio and Coder.
Will Alibaba release Qwen4-Max weights?
No. Alibaba did not release the model weights but said smaller Qwen4 series models will be released as open weights later, without giving sizes or dates.