Tuesday, November 17, 2026 · Qwen

From Issue 105 (2026-11-17) · 11 stories in this issue

04 MODEL

❯ Alibaba launches Qwen4-Max with 2.4 trillion parameters and visual input

A new flagshipAlibaba has launched Qwen4-Max, a flagship model with 2.4 trillion total parameters, native vision capabilities and a one-million-token context window. It is designed for coding, agent tasks, workplace use and research.

Parameters used on demandThe model uses a mixture-of-experts architecture, activating only selected parts for each token. Alibaba says it activates about 71 billion parameters per token, 25% fewer than Qwen3.8-Max despite the same total parameter count, reducing the computing required to run it.

Performance and costOn Alibaba’s own Work360 benchmark, Qwen4-Max performed better overall than its predecessor at high reasoning effort while using 35% fewer tokens and reducing average task cost by about 60%. Real-world cost depends on the task and pricing. Independent evaluator Artificial Analysis also found a higher score and lower cost per task on its own index.

Developer accessQwen4-Max is available through QwenCloud’s API and subscription plan, as well as QwenWork and QwenStudio. Alibaba says smaller open-weight Qwen4 models will follow; they are separate products from the current flagship.

▮ SIGNALAs companies use AI for long documents and multi-step work, competition increasingly depends on the computing needed to complete each task.