Alibaba launches Qwen4-Max, its most advanced AI model that cuts task costs by 60%
by Paul

Alibaba launches Qwen4-Max, its most advanced AI model that cuts task costs by 60%

Alibaba has officially released Qwen4-Max, its most advanced AI model to date, positioning the new flagship as a teammate built for real-world work. The model keeps the same 2.4 trillion parameters as its predecessor, Qwen3.8-Max, but uses a more efficient mixture-of-experts design that activates only 71 billion parameters per token, roughly 25% fewer than before. According to Alibaba's internal Work360 benchmark for workplace tasks, that efficiency translates into a 60% drop in average cost per task, from $4.43 to $1.67. Qwen4-Max features a 1-million-token context window, accepts both text and image input, and is available now through QwenCloud with input priced at $0.075 per million tokens.

The company paired the performance pitch with an unusually heavy emphasis on safety. Alibaba said it deliberately filtered Qwen4-Max's training data to strip out offensive cyberattack capabilities while preserving defensive skills such as vulnerability detection, and it added dedicated datasets and post-training measures to mitigate chemical, biological, radiological and nuclear risks. Alibaba also showcased the model's scientific chops, citing a continuous 48-hour run in which it discovered an experimentally validated coolant molecule for next-generation semiconductor chips, along with autonomous research on Alzheimer's disease. Although Qwen4-Max itself is closed-source, Alibaba reaffirmed that smaller, open-weight versions of the Qwen4 series will be released to global developers soon, ahead of its planned Qwen 5 model, which it says will be scaled to between 5 and 10 trillion parameters.

Early independent appraisals suggest the model is competitive but not untouchable. Evaluators at Vals AI ranked it the top Chinese model on their index shortly after launch, and it topped four of sixteen charted benchmarks, but Artificial Analysis noted it scores slightly behind Moonshot's Kimi K3.1 Pro on aggregate intelligence while offering lower cost per task, and it trails rivals on agentic terminal and real-world coding tests.

Paul
Written by Paul
Follow
Add as a preferred source on Google
Qwen iconQwen
  88
  • ...

Qwen is an AI chatbot interface for interacting with Alibaba Qwen's series of models, known for their extensive context windows and open-source nature. With a 4.7 rating, Qwen offers features like ad-free usage, dark mode, and no registration requirement. These attributes, along with generous free usage limits, contribute to its reputation for reliability.

Other apps mentioned in this article

  • Qwen Code icon
    Qwen Code
    AI Coding Assistant, Large Language Model (LLM) Tool

No comments so far, maybe you want to be first?