AI
Alibaba Unveils Qwen4-Max AI Model With 60% Cost Reduction
The 2.4 trillion-parameter model features enhanced safety guardrails and delivers improved reasoning at significantly lower inference costs.
Key takeaways
- Alibaba's Qwen4-Max AI model reduces average task costs by 60% through more efficient parameter activation.
- The model activates only 71 billion of its 2.4 trillion parameters per token, 25% fewer than its predecessor.
- Alibaba implemented dedicated security measures to mitigate chemical, biological, radiological, and nuclear risks.
- While Qwen4-Max is closed-source, smaller open-weight versions will be released for developers.
- The model is priced at 7.5 U.S. cents per million tokens through QwenCloud.
Alibaba Group has released Qwen4-Max, its most advanced artificial intelligence model to date, featuring substantial cost reductions and enhanced security measures designed to address growing concerns about AI safety.
The Hangzhou-based technology conglomerate's new model reduces average task costs by 60 percent compared to previous versions, according to internal benchmarks evaluating performance across various work tasks, The South China Morning Post first reported. The model is now available through QwenCloud at a price of 7.5 U.S. cents per million tokens.
Technical architecture and efficiency gains
Qwen4-Max comprises 2.4 trillion parameters, matching the size of its predecessor Qwen3.8-Max. However, the new model achieves greater efficiency by activating only 71 billion parameters per token—25 percent fewer than the previous generation. This architectural approach enables improved reasoning capabilities while simultaneously reducing computational costs and inference expenses.
The efficiency improvements represent a significant development in large language model design, where reducing the number of active parameters without sacrificing performance has become a key engineering challenge as models scale.
Security measures and open-source strategy
Alibaba has implemented stringent cybersecurity guardrails for Qwen4-Max, including dedicated datasets and post-training enhancements specifically designed to mitigate chemical, biological, radiological, and nuclear (CBRN) risks. These safety measures address mounting global concerns about AI-enabled cyberattacks and the potential misuse of powerful AI systems.
While Qwen4-Max itself is a closed model, Alibaba reaffirmed its commitment to open-source AI development. The company stated that smaller-scale, open-weight versions of the Qwen4 series will be released for global developers in the near future. This dual approach reflects an attempt to balance the benefits of open-source collaboration with the need for controlled deployment of the most capable models.
The launch comes as Chinese AI companies navigate questions about whether open models could increase security risks compared to the closed-source approaches favored by U.S. competitors such as OpenAI and Anthropic.
Why it matters
The 60 percent cost reduction in Qwen4-Max could accelerate enterprise AI adoption by making advanced language model capabilities more economically viable at scale. For organizations evaluating AI infrastructure investments, lower inference costs directly impact the total cost of ownership for production deployments. Alibaba's explicit focus on CBRN risk mitigation also signals that major AI providers are taking concrete steps to address safety concerns that have largely remained theoretical—a shift that may influence regulatory frameworks and industry standards for responsible AI development.
Market reaction
U.S.-listed shares of Alibaba declined more than 1 percent in overnight trading following the announcement, suggesting investors may be focused on broader market factors rather than the technical achievements of the new model.
Details of the launch were first reported by The South China Morning Post.