AI News

Alibaba's Tongyi Qianwen releases Qwen4-Max, adopting a new MoE architecture and supporting a 1M token context

Tongyi Qianwen / QwenUpdated Nov 16 20:43

AI summary

Alibaba's Tongyi Qianwen has released Qwen4-Max, based on a new MoE architecture, with native visual support and a context window of 1M tokens. The company says that in internal evaluations it surpasses Qwen3.8-Max while using up to about 50% fewer tokens, with average cost per task dropping by more than 60%; it scores 52.69 on Work360, about 10% higher than Qwen3.8-Max. The API is now live on QwenCloud, at $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit cache, and it has been announced that the weights of smaller Qwen4 models will be open-sourced.

Why read this

AI score 77

The original text provides the new MoE architecture, the 1M context, and an item-by-item token cost comparison, so readers can use it to assess changes in cost-effectiveness for long-task scenarios.

Topics

Tags

Body

3D understanding and design: from a rough blockout to a cinematic cut. https://t.co/29WMBqh7CC

Source

Tongyi Qianwen / Qwen