Alibaba Group Holding has launched Qwen4-Max, describing it as its most advanced artificial intelligence model to date, with strict cybersecurity guardrails and a claimed 60 per cent reduction in cost per completed task.
The metric choice is telling. Labs increasingly report cost per task rather than cost per token, because agentic workloads consume variable numbers of tokens to finish a job; a model that reasons more efficiently can be cheaper even at the same per-token price.
The guardrail language matters for enterprise sales in China, where content and data rules are explicit and buyers need documented controls before deployment.
Alibaba is simultaneously keeping its open-weight strategy for smaller models, a dual track that lets it seed developer habits at the low end while monetising frontier capability through the API.
The claim to test externally is the 60 per cent figure: task-level cost depends heavily on how a benchmark defines a task, and vendor numbers rarely survive independent reproduction unchanged.
