Alibaba launches Qwen4-Max AI model with strict safety upgrades
Alibaba Group Holding officially unveiled its new flagship artificial intelligence model Qwen4-Max in China today alongside major cybersecurity guardrails.

Model specifications and performance gains
The new system utilizes a sparse Mixture-of-Experts architecture with a total of 2.4 trillion parameters. Internal testing found no evidence of significant high-risk capabilities within the design. Alibaba stated the model activates only 71 billion parameters per token during operation. This efficiency cuts average task costs by roughly 60 percent compared to its predecessor.
Safety measures and data filtering
The firm clearly removed offensive cyber tools from its training dataset while reducing serious threats. These upgrades target chemical, biological, radiological and nuclear threats as well as general cybercrime misuse. Alibaba said it heavily filtered inputs to prevent the generation of hazardous material instructions. This approach addresses growing global concern surrounding rogue AI and potential loss-of-control risks.
Commercial rollout and capabilities
Qwen4-Max features a 1 million token context window for handling complex cross-modal reasoning tasks. The model natively handles text, image and video comprehension across various professional workloads. It is currently rolling out via Alibaba QwenCloud and Model Studio platforms for users. Pricing is designed to challenge domestic rivals like DeepSeek and Moonshot as well as US giants.
All 2 outlets report consistently
No differences in figures, names or dates across these reports.
- The News International
- South China Morning Post
Reported by
2 independent outlets. Headlines as published. Links open the original report.