Alibaba releases Qwen4-Max with 1M-token context and 60% lower cost per task

·2026.11.16 14:04

Key point

The new MoE model supports 1M tokens and reduces average cost per task by over 60% compared to Qwen3.8-Max.

Details

  • Alibaba releases Qwen4-Max, a new MoE architecture model featuring native vision and a 1M-token context window.
  • Internal evaluations show Qwen4-Max outperforms Qwen3.8-Max while using up to 50% fewer tokens and reducing average cost per task by over 60%.
  • Pricing is set at $1.5/M input tokens, $4.5/M output tokens, and $0.075/M implicit caching tokens, with implicit caching costs down 70% from the previous version.
  • The API is live on QwenCloud, and the model is available on Qwen Studio, QwenWork, and Qoder.
  • Smaller Qwen4 models with open weights are planned for future release.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.