Alibaba Unveils Qwen4-Max Flagship AI With 60% Lower Task Costs
On Monday, Alibaba introduced Qwen4-Max, its newest flagship artificial intelligence model, asserting that it lowers average task costs by approximately 60 percent while enhancing protections against cybersecurity threats and various risks. The company stated that the model surpasses its predecessor, Qwen3.8-Max, in handling complex professional tasks while utilising 35 percent fewer tokens in high-effort reasoning mode. According to the reports, Qwen4-Max boasts 2.4 trillion parameters, incorporates native vision capabilities, and features a context window of one million tokens. Alibaba described it as the “most capable model in the Qwen family to date”. Qwen4-Max is constructed on an enhanced sparse Mixture-of-Experts architecture, enabling it to engage only a subset of its total parameters for each token processed, according to the company. Although the model comprises 2.4 trillion total parameters, it activates 71 billion parameters per token, which is 25% fewer than Qwen3.8-Max at the same overall scale, according to Alibaba. The company reported that Qwen4-Max surpassed Qwen3.8-Max on its proprietary Work360 benchmark, which assesses performance across 721 specialised fields.
In a state of heightened analytical engagement, the model demonstrated a reduction of 35% in token usage while simultaneously achieving a decrease of approximately 60% in the average cost per task. Qwen4-Max secured the sixth position on Code Arena, a platform that evaluates coding skills in real-world scenarios. Alibaba reported robust outcomes on various metrics related to agent workflows, workplace productivity, and visual reasoning, including Agents’ Last Exam, Vals Index, JobBench, and Chartography. The model is structured to address tasks in 3D modelling, game development, precision manufacturing, finance, law, and healthcare. Alibaba stated that Qwen4-Max additionally facilitates cloud resource scheduling. In benchmark testing, the scheduling capabilities led to a reduction in data-annotation model call costs by an average of 9%, while also enhancing offline CPU rescheduling speeds by over 14 times.
In the realm of automotive manufacturing, the model serves as a valuable tool for optimising production-line planning, enhancing part design, and ensuring machining verification. During a company demonstration, manual workpiece programming time was reduced by 95 percent, from approximately 475 minutes to just 24 minutes, by pinpointing excessive cutting and inadequate material removal prior to the commencement of machining. Alibaba also reported using Qwen4-Max for an experiment involving Alzheimer’s research. The model handled tasks covering target research, molecular design and computational evaluation. During the workflow, it autonomously executed 454 tool calls and 3,239 metric evaluations to design and optimise 24 candidate molecules, according to the company. Alibaba said it incorporated security checks into software engineering training alongside functionality tests to address vulnerabilities such as code injection, memory-safety flaws and broken access controls. The company additionally refined the training data based on the intent of the tasks.
It retained defensive cybersecurity material used for vulnerability detection and remediation while removing content related to offensive cyberattacks. Additionally, Alibaba has unveiled specialised safety datasets and enhancements following training to bolster defences against chemical, biological, radiological, and nuclear threats. According to source, Alibaba said its internal testing had found no evidence of significant high-risk capabilities in the model. The company also stated that it had bolstered protections against potential misuse in domains such as cybercrime and hazardous materials. Qwen4-Max is accessible through Alibaba’s QwenCloud platform via its Token Plan and application programming interface. Access is also available through QwenWork and QwenStudio. Alibaba has set the price for input tokens with an implicit cache hit at $0.075 per million tokens. The company announced that smaller, open-weight models from the Qwen4 series would be made available to developers globally.








