Models New
Alibaba releases Qwen4-Max, its new flagship model, through its paid API with a one-million-token context
Published:
Alibaba has released Qwen4-Max, the new flagship of its Qwen family, Reuters reported early on Monday Paris time. QwenCloud's model page lists it as a multimodal model that reads text, images and video and writes text, with a 1M-token context and up to 393,000 tokens of output, priced at $1.50 per million input tokens and $4.50 per million output tokens.
Alibaba told Reuters that Qwen4-Max outperforms Qwen3.8-Max, released in August, while using fewer tokens, and that its training included security checks alongside coding tests to reduce flaws such as code injection and broken access controls. The company said its internal testing found no evidence of significant high-risk capabilities.
According to Sulat AI & Tech News, which summarised Qwen's launch blog, the model is a mixture-of-experts design with a 2.4-trillion-parameter backbone and 71 billion parameters active per token, and Qwen's own Work360 evaluation puts the average cost per task at $1.67, down from $4.43 for Qwen3.8-Max. Qwen says smaller Qwen4 models will follow as open weights, without sizes or dates.
Caveat. Performance, token savings and safety findings are Alibaba's own claims, mostly on its in-house Work360 benchmark; they were not independently verified. Qwen's launch blog is a script-rendered page that the tools used here could not read, so the parameter counts come from Sulat's summary of it.
Sources