Alibaba Says New AI Model Cuts Task Costs by 60 Percent

Alibaba Group Holding has unveiled its most sophisticated AI model to date, Qwen4-Max, equipped with stringent cybersecurity measures, as the tech giant navigates safety while maintaining its dedication to open sourcing in light of increasing worldwide worries regarding AI-driven cyberattacks.
The announcement underscores an increasing emphasis on safety within China's AI sector, as Alibaba noted it deliberately screened the model's training data to eliminate harmful cyberattack features while preserving functions beneficial for cyber defense like vulnerability detection.
The company based in Hangzhou also unveiled specific data sets and post-training improvements to address chemical, biological, radiological, and nuclear (CBRN) threats.
Also Read: EcoWorld, NS Corp to Develop Industrial Water Supply Solutions
Alibaba also emphasized the model’s advancements in scientific studies, including a continuous 48-hour experiment to identify an experimentally confirmed coolant molecule for future semiconductor chips, along with self-directed research on Alzheimer’s disease. There is intense discussion regarding whether open-weight models, preferred by Chinese firms, may increase CBRN risks more than the closed-source models supported by US competitors like OpenAI and Anthropic.
Certain specialists have cautioned that AI technologies might ultimately become potent enough to assist harmful individuals or nations in creating CBRN weapons. Although Qwen4-Max is a proprietary model, Alibaba reiterated its commitment to open-source by announcing that smaller, open-weight versions of the Qwen4 series will be made available to worldwide developers shortly.
Also Read: Oman Awards Green Hydrogen Projects in Duqm, Launches Auction
Alibaba's Max series of models were previously closed-source until Qwen 3.8 was released as open source in August. The Max series is additionally the largest model of the company in terms of parameter size. Nonetheless, the company emphasized its successes in enhancing the cost-effectiveness of its newest model for users.
Qwen4-Max features 2.4 trillion parameters, identical in size to the earlier Qwen3.8-Max, but it engages only 71 billion parameters for each token – 25 percent fewer than its forerunner. This enables it to provide enhanced reasoning abilities while reducing inference expenses, lowering the average cost per task by 60 percent, as per Alibaba, referencing an internal benchmark named Work360, which assesses model performance on different work assignments.
The model is now accessible through QwenCloud, with input costs established at 7.5 US cents for every one million tokens.
Alibaba has concentrated its AI initiatives primarily on seizing the enterprise AI sector and fostering an open-source community to surpass local competitors such as DeepSeek and Z.ai, along with those from the US.
Liu Dayiheng, leader of Alibaba’s Qwen large language model team, revealed a roadmap detailing intentions for the Qwen 4, Qwen 4.5, and Qwen 5 series in September, while highlighting the team’s dedication to the open-source community earlier this month at the World Internet Conference in Wuzhen.
Also Read: UAE to Establish Mideast First Hot Air Balloon Museum With Japan
The firm has announced that its upcoming model, Qwen 5, will be enhanced to feature between five trillion and 10 trillion parameters.

