What happened
Alibaba has launched Qwen4-Max, the largest AI model it has built to date, Trending Topics reported. The mixture-of-experts system holds 2.4 trillion parameters, of which 71 billion are activated for each token, and it supports a context window of one million tokens. It reads images as well as text, performs reasoning, and is positioned by the Qwen team as a round-the-clock collaborator for work like 3D modelling, games and legal reviews.
On the Intelligence Index from Artificial Analysis, the model scored 43 points. It came ninth of 10 leading models compared head to head, behind systems from Anthropic, Google, OpenAI and Moonshot AI. Within its model class it ranked 38th of 81, slightly above the median score of 41, leaving Moonshot's Kimi K3.1 Pro as the strongest Chinese entry. Alibaba's own published comparisons also show rivals ahead in agentic coding.
Qwen4-Max is proprietary. Its weights cannot be downloaded, and access runs through the QwenCloud API and the QwenWork, Qoder and QwenStudio apps. Artificial Analysis lists pricing at $1.50 per million input tokens and $4.50 per million output tokens, with cached inputs discounted by 95 percent. According to the Qwen team, smaller Qwen4 models sharing the same architecture and training techniques will be released as open source.
Context
Since Qwen2.5-Max, Alibaba has kept its Max models restricted to its own cloud, while its open models built a large developer base and recently prompted Tencent to follow. Jensen Huang has said Chinese open models are used by 80 percent of US start-ups. Alibaba has not disclosed the hardware used to train Qwen4-Max. This is sensitive after a forensic analysis linked banned Nvidia chips to recent models from several Chinese developers.
Обзор
Неделя в технологиях в одном письме.
Материалы, которые стоят вашего времени, каждую неделю. Отписаться можно в любой момент.
