Alibaba announced on November 16, 2026 the release of Qwen4-Max, the most capable model in the Qwen family to date. With a Mixture-of-Experts (MoE) architecture of 2.4 trillion total parameters and only 71 billion active per token, the model accepts text, images and video and offers a 1 million token context window.
Efficiency and performance
According to Alibaba, Qwen4-Max outperforms its predecessor Qwen3.8-Max on the proprietary Work360 benchmark covering 721 specialized fields, while using about 35% fewer tokens in High effort mode and reducing average cost per task by around 60%. The company highlights gains in agentic tasks, coding and visual reasoning.
The model is available via API on QwenCloud, Token Plan, Qwen Studio, QwenWork and Qoder. Listed prices are $1.50 per million input tokens and $4.50 output, with implicit cache at $0.075. Alibaba plans to release smaller open-weight versions of the Qwen4 series.
Safety and applications
The company embedded security checks in software engineering training and filtered data to reduce risks in cybersecurity and sensitive areas. The model is positioned for professional work, creativity, scientific research and long-running agentic workflows.
Sources: Alibaba Cloud, Reuters, MarkTechPost.
This article was produced with the assistance of artificial intelligence and editorially reviewed by Grok.
By Grok