Briefs · Taipei time (GMT+8)

Alibaba's Qwen launches Qwen4-Max, with its API now on QwenCloud

一句重點:Alibaba's Qwen says Qwen4-Max outperforms Qwen3.8-Max (🗣 self-reported), uses up to about 50% fewer tokens, and has lower prices.

發生什麼事

@Alibaba_Qwen says Qwen4-Max uses a new mixture-of-experts (MoE) design, handles vision natively, and has a context window of 1 million tokens. The company reports better performance with professional tools, longer task execution, and the quality of its results. It lists real-world work such as 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier science. Qwen says the model uses up to about 50% fewer tokens than Qwen3.8-Max (🗣 self-reported). In the company's internal evaluation, average cost per task fell by more than 60% (🗣 self-reported; not tested by this site).

The API is available on QwenCloud, and the model can also be tried in Qwen Studio, QwenWork, and Qoder. Pricing is $1.5 per million input tokens, $4.5 per million output tokens, and $0.075 per million tokens for implicit caching. All three rates are below Qwen3.8-Max; the caching rate is 70% lower.

01Sources

  • @Alibaba_Qwen: Alibaba's official Qwen account, posted on 2026-11-16 at 02:01 UTC.