Skip to content
Book a boothGet tickets

Alibaba Unveils Qwen4-Max AI Model with 2.4 Trillion Parameters

Alibaba has introduced Qwen4-Max, an AI model with 2.4 trillion parameters and a 1 million-token context window. Designed for coding, scientific research and enterprise applications, the model reportedly reduces average task costs by 60% and is accessible…

Alibaba Unveils Qwen4-Max AI Model with 2.4T Parameters

Chinese technology company Alibaba has unveiled its Qwen4-Max artificial intelligence model, featuring 2.4 trillion parameters and a context window of 1 million tokens. The company said the model, developed for coding, scientific research, and enterprise workflows, reduces average task costs by approximately 60% compared with its predecessor. The model is available through QwenCloud.

Announced by Alibaba Cloud in Hangzhou, China, Qwen4-Max is positioned as the most capable model in the company's Qwen AI family. The model can process images directly and is designed to analyze lengthy documents, develop software, perform complex tasks, and automate professional workflows.

According to results released by Alibaba, Qwen4-Max ranked sixth on the Code Arena leaderboard, which evaluates performance on real-world software development tasks. The company also reported strong performance in autonomous task execution by AI agents, business process efficiency, and visual reasoning.

Qwen4-Max Reduces Processing Costs by 60%

According to Alibaba's Work360 benchmark, developed to evaluate tasks across 721 professional domains, Qwen4-Max achieved higher overall performance than its predecessor, Qwen3.8-Max. Company data indicates that the model uses 35% fewer tokens at high reasoning levels, reducing the average cost per task by approximately 60%.

The model uses a Mixture-of-Experts (MoE) architecture. Rather than activating all parameters for every operation, this architecture activates only the relevant expert components. Qwen4-Max uses approximately 71 billion of its total 2.4 trillion parameters per token. This represents a 25% reduction compared with the previous model of a similar scale. These performance and cost figures are based on benchmarks published by Alibaba.

Qwen4-Max Used in Scientific Research and Industry

Alibaba stated that Qwen4-Max can be used in areas including 3D design, game development, video production, finance, law, healthcare, and precision manufacturing. In an automotive manufacturing application shared by the company, the model reduced the time required to prepare component machining programs from approximately 475 minutes to 24 minutes.

This represents a 95% reduction in manual programming time. In scientific research, Qwen4-Max was used in an Alzheimer's disease experiment involving 454 tool calls and 3,239 measurement evaluations, contributing to the design and optimization of 24 candidate molecules. Alibaba also reported that the model family was involved in a 48-hour research process focused on chip cooling technologies.

Alibaba Strengthens Security Measures in Qwen4-Max

Alibaba announced that cybersecurity and software security controls were integrated into the training process during development of the new model. The company stated that it implemented measures designed to reduce risks associated with malicious code generation, exploitation of software vulnerabilities, and unauthorized access.

Alibaba reported that the model's success rate in secure code generation evaluations increased from 8% in the previous version to 14.7%. The company also announced improvements to safeguards against chemical, biological, radiological, and nuclear risks. These security results have not been independently verified.

Qwen4-Max Becomes Available Through Cloud Platforms

Qwen4-Max has been made available on Alibaba's QwenCloud platform through Token Plan subscriptions and API access. An application programming interface (API) enables software developers to integrate the model into their own applications.

The model is also accessible through the QwenWork and QwenStudio platforms. Alibaba announced a price of $0.075 per 1 million tokens for cached input data. The company also stated that it plans to release smaller-scale, open-weight models in the Qwen4 series for developers worldwide. However, no specific release date has been announced for these models.

Read more on

About the author

Uğur Gürbes

278 stories on WORLDEF since February 2026.

More from Uğur Gürbes →

The week in commerce, before your competitors.

1,497 stories since January 2025, distilled into three newsletters. Pick the ones you want.