Alibaba Group has introduced Qwen4-Max, its latest flagship artificial intelligence model, claiming improvements in coding, cybersecurity, professional workflows and scientific research. The Chinese technology company says the model can compete with leading US-developed AI systems, including those from OpenAI and Anthropic.
Built on Alibaba's new Qwen4 architecture, the model uses a mixture-of-experts design with a reported 2.4 trillion parameters, of which 71 billion are activated for each token processed. This approach allows the system to use selected components of the model for individual tasks rather than activating its entire architecture.
Qwen4-Max supports a context window of one million tokens, enabling it to process lengthy documents and large volumes of information within a single interaction. It also includes native vision capabilities for interpreting visual content alongside text.
According to Alibaba, the model is designed for applications ranging from software development and debugging to financial document analysis, game development and scientific research.
The company has also released benchmark results comparing Qwen4-Max with competing AI models. In its internal DeepSWE v1.1 coding evaluation, Qwen4-Max recorded a score of 75.7, compared with 75.2 for OpenAI's GPT-6.1-Sol and 74.2 for Anthropic's Opus 5.5.
However, these results are based on company-reported testing and do not establish that Qwen4-Max performs better across all tasks or independent evaluations.
Alibaba has also highlighted improvements in computational efficiency. The company claims Qwen4-Max uses up to 35% fewer tokens than its predecessor, Qwen3.8-Max, on its Work360 benchmark and 49% fewer tokens on Terminal-Bench 4.0.
Average task expenditure on Work360 reportedly declined by 62%, from $4.43 to $1.67, according to Alibaba's internal measurements.
In cybersecurity, the model is designed to identify software vulnerabilities and recommend corrective measures while restricting responses that could facilitate cyberattacks. Alibaba says it has incorporated additional safeguards addressing potential misuse involving chemical, biological and nuclear threats.
Unlike some earlier models in the Qwen series, Qwen4-Max is closed-source, meaning developers cannot independently download and operate its underlying model weights. Alibaba has indicated that smaller models within the Qwen4 family will be released as open-source offerings.
The model is available through QwenCloud, QwenWork, Qoder and QwenStudio. Alibaba has also introduced a feature called Live Delivery, which allows users to receive an initial working version of a project and refine it through subsequent instructions.
Cached-input API pricing starts at 0.5 yuan per million tokens, although overall usage costs depend on the type and volume of processing.
The launch adds another model to the expanding enterprise AI market, where providers are competing on coding performance, operating costs, multimodal capabilities and integration into business workflows.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.