Alibaba’s rate card
Alibaba opened Qwen4-Max on the QwenCloud Token Plan and API. The new rates are lower than the previous Max rates. The change to cached input is particularly large. Alibaba says smaller Qwen4 models will be open, while Max remains available through the hosted service.
Alibaba reports a lower bill and higher score on its own Work360 benchmark. This is a vendor claim. Artificial Analysis also measured an improvement over the previous Max on its Intelligence Index.
What the task costs
Qwen4-Max read less input than its predecessor on the independent index, and the input itself was cheaper. Cache writes still accounted for the largest part of its bill. Claude Fable at low read markedly less and scored higher for a similar cost. Sol at medium was considerably cheaper and scored higher.
Where to try it
Vals AI put Qwen4-Max higher than the previous Max. Its legal results were stronger than Kimi’s on the cited Vals tasks, while terminal work was a weaker fit. Only extra-high had independent measurements at the time. A trial for front ends or long document jobs can be worthwhile if the wait fits the workload.
