Qwen4-Max packs 2.4 trillion parameters and cuts task costs by 62%
Alibaba's Qwen team unveiled Qwen4-Max, a 2.4-trillion-parameter model built on a new architecture with native vision and a one-million-token context window. It activates 71B parameters per token and adds 328B N-gram embedding parameters. The model beats its predecessor across coding, professional work, and long-horizon tasks while using fewer tokens, and average cost per task on the Work360 benchmark falls from $4.43 to $1.67. It also drives long-running agents and end-to-end workflows in 3D modeling, game development, law, finance, and healthcare.
In this showcase, the manual effort required to program a single workpiece fell from approximately 475 minutes to 24 minutes, a reduction of about 95%.