Alibaba releases Qwen4-Max
Alibaba released Qwen4-Max, the company's most advanced model to date with 2.4 trillion total parameters, more efficiency and improvements in coding, workflows and productivity.
Qwen4-Max is worth watching since Alibaba's latest model outperforms Qwen3.8-Max for various tasks while using 35% fewer tokens. The average cost per task with Qwen4-Max is 60% lower than its predecessor.
Alibaba said smaller, open-weight models from the Qwen4 family will be released soon.
The launch of Qwen4 comes as the leader in the US open model race, Nvidia's Nemotron family, may be derailed by corporate issues with Nvidia, which is looking for a permanent CEO after the Jensen Huang scandal.
Alibaba and the Qwen team cited numerous benchmarks indicating that Qwen4-Max can compete well against open models as well as proprietary ones.
The other interesting thread to ponder is how Chinese open models are increasingly being sanitized and launched under other projects such as Nous' Hermes models. Nous Research launched Arete-27, a Qwen3.8-27B post-trained model that reinforces western values.
Nous aims to de-risk open models.
In the end, the Qwen4-Max launch is likely to spawn a series of derivatives.