Just came across this:
Alibaba has officially launched Qwen4-Max
On November 11, it was still just a limited-time preview
Today, it officially went live on the API
The new architecture is MoE
It comes with visual understanding
1 million-token context window
In the official Work360 test
The score rose from 47.74 to 52.69
Average cost per task fell from $4.43 to $1.67
A drop of over 60%
Token usage in coding tests also fell by about half
Current listed prices for the domestic API:
10 yuan per million input tokens
30 yuan for output
0.5 yuan for cached input
The official claim is that cached input is 70% cheaper than the previous generation
One official case is pretty wild:
Programming for auto parts machining
Cutting the time from about 475 minutes by hand to 24 minutes
They also say smaller Qwen4 models will be open-sourced later
For now, just note the numbers
Now that it’s officially launched, let’s see if it holds up
Alibaba has officially launched Qwen4-Max
On November 11, it was still just a limited-time preview
Today, it officially went live on the API
The new architecture is MoE
It comes with visual understanding
1 million-token context window
In the official Work360 test
The score rose from 47.74 to 52.69
Average cost per task fell from $4.43 to $1.67
A drop of over 60%
Token usage in coding tests also fell by about half
Current listed prices for the domestic API:
10 yuan per million input tokens
30 yuan for output
0.5 yuan for cached input
The official claim is that cached input is 70% cheaper than the previous generation
One official case is pretty wild:
Programming for auto parts machining
Cutting the time from about 475 minutes by hand to 24 minutes
They also say smaller Qwen4 models will be open-sourced later
For now, just note the numbers
Now that it’s officially launched, let’s see if it holds up

