Alibaba launches Qwen4-Max and promises 60% cut in AI costs for businesses

2–3 minutes

Chinese tech giant Alibaba announced on Monday, 16th, a new version of its artificial intelligence model, Qwen4-Max. According to an internal assessment, this model can reduce the average cost per task by around 60% compared to the previous generation. The results were measured in the Work360 test, which simulates professional activities across 721 specialized fields.

Cost reduction and operational efficiency

Qwen4-Max retains the same parameter base as its predecessor, with 2.4 trillion parameters, but only activates 71 billion per token—a 25% decrease. The architecture selects specific parts of the model for each processing step, rather than using the entire structure for every response. This approach resulted in a 35% reduction in token usage during high-effort reasoning mode.

For companies performing large-scale repetitive tasks, this difference is significant. The total cost of an application depends on both the fee charged and the amount of information processed to achieve the desired outcome.

New security measures

Alibaba has also enhanced the model’s protections against offensive uses and cybersecurity threats. The company separated training materials by purpose, preserving content aimed at identifying and correcting vulnerabilities while removing material related to offensive attacks. Security checks were integrated into programming training, and barriers against chemical, biological, radiological, and nuclear risks were strengthened.

Although the developer has announced these measures, there is no guarantee that the system will be immune to misuse.

Market strategy and model access

Qwen4-Max is a closed model, but Alibaba plans to release smaller versions of the Qwen4 series with open weights, allowing developers to adapt the system to their needs. This strategy combines selling access to the main model with offering customizable solutions—a key aspect of Chinese companies’ competition in the global AI market.

Access to the model is provided through QwenCloud, as well as developer interfaces and products like QwenWork and QwenStudio. In Brazil, Alibaba Cloud launched a cloud region with two data centers in August, but the local presence does not guarantee immediate availability of Qwen4-Max in the Brazilian infrastructure.

Frequently Asked Questions

  • Does Qwen4-Max cost 60% less? Alibaba reports an average cost reduction per task in its Work360 test, but this does not equate to a universal discount for all users.
  • What are tokens? They are units of information processed by the model and used to charge for AI services.
  • Is Qwen4-Max open source? No. Alibaba plans to release smaller versions of the Qwen4 series with open weights.
  • How does the model combat offensive uses? The company says it has filtered training data to preserve cybersecurity and remove content related to attacks.
  • Where can I access the model? Alibaba provides access through QwenCloud and products like QwenWork and QwenStudio.
  • Is the model already available in Brazilian infrastructure? The presence of data centers in Brazil does not confirm the availability of Qwen4-Max in this region.