The same Qwen model costs less from Beijing than from Singapore
Alibaba publishes Model Studio prices separately for each deployment, and the same model is not the same price from Singapore and from Beijing[1].
Per million tokens, qwen3.8-max is $2 input and $6 output on the international deployment in Singapore, and $1.65 and $4.951 in Beijing[1]. That is 17.5 percent off both numbers. qwen3.8-flash is $0.15 and $0.47 against $0.113 and $0.382[1].
qwen3.7-plus steps by prompt length as well[1]. From Singapore, input is $0.4 up to 256K tokens and $1.2 above it, and output is $1.6 and $4.8[1]. From Beijing the same steps are $0.276 and $0.826 on input, and $1.101 and $3.301 on output[1].
The tier is set by the input length of the whole request, and every token in the request is billed at that tier's price[1]. A limited-time 20 percent discount is shown on qwen3.7-plus in both places[1].
Cache hits are billed at 10 percent of the input price, and creating an explicit cache at 125 percent[1]. Where a model supports batch calls, input and output are both half price[1].