- Introduction to the new technical release
- Innovative architecture and selective parameter activation
- Superior performance and competition in the global market
- Revenue-sharing strategy and commercial transformation
- Developer platforms and artificial intelligence agents
- Frequently asked questions
Introduction to the new technical release
In a prominent strategic move that redraws the global technology landscape and highlights intense competition in the advanced tech sector, Alibaba Group Holding has unveiled its latest artificial intelligence innovation: a multimodal model named Qwen 3.8 Max. The model features immense computational and cognitive capabilities, boasting approximately 2.4 trillion parameters, placing it among the most advanced and massive models globally. This announcement marks a turning point and a strong return by the Chinese tech giant toward an open-source model strategy, following a period of conservatism earlier this year during which the company kept many of its leading releases as closed, exclusive products. Alongside the launch of this massive model, the company introduced an innovative revenue-sharing model targeted specifically at commercial users, and opened an advanced developer platform. These three-dimensional initiatives reflect the company’s persistent efforts to build an integrated and comprehensive artificial intelligence ecosystem, in addition to achieving sustainable financial returns from its open-weight strategy for developers and enterprises.
Innovative architecture and selective parameter activation
The Qwen 3.8 Max model relies on a highly sophisticated and complex architectural structure known in scientific circles as the Mixture of Experts. This innovative technique enables the activation of only a specific part of the neural network during processing, with the model activating just 95 billion real-time parameters out of its total 2.4 trillion. This selective activation effectively keeps operating costs extremely low and significantly cuts response times compared to traditional models that require full network activation at such a massive scale. Furthermore, the model supports an exceptional, massive context window capable of handling up to one million tokens simultaneously, giving it unprecedented capability to process and analyze very long texts, as well as understand complex images and high-resolution videos with high speed. The model is now globally available through APIs on Alibaba Cloud’s Model Studio platform, with official commitments to release open weights to the public and developers in later stages to maximize its utility across various projects.
Superior performance and competition in the global market
Qwen 3.8 Max proved its merit and high efficiency immediately upon release, achieving stunning results on the independent benchmarking platform LMSYS Chatbot Arena. The model immediately topped the rankings to become the highest-rated Chinese model in text task processing, although it still trails slightly behind leading advanced releases from American firm Anthropic in this competitive arena. In computer vision and image analysis, the model achieved a major milestone by ranking second globally, demonstrating its superior capability in processing visual data. The model closely approaches the scale and capabilities of its prominent domestic rival, Moonshot AI’s Kimi K2, which launched successfully last month with 2.8 trillion parameters. This clearly reflects intensifying competition in the Chinese and global markets within this fast-growing vital sector, underscoring the ability of Chinese developers to deliver strong, high-performance alternatives.
Revenue-sharing strategy and commercial transformation
In a major strategic shift in monetization policy, Alibaba Group plans to introduce new requirements obligating major commercial users and enterprises heavily relying on Qwen 3.8 Max’s open weights to share a portion of the revenue and profits generated from such usage. According to a Reuters report citing two people familiar with these confidential plans, the exact profit-sharing percentage is still under discussion and review by senior management. This move follows a similar approach by Moonshot AI, whose new model licensing requires partners generating annual sales exceeding $20 million to enter into a special, detailed commercial agreement involving profit sharing of up to 30 percent. This trend represents a fundamental change in Alibaba’s policy, which historically permitted the use of its open-source models within corporate data centers without any financial fees or restrictions. Economic analysts and experts have described this approach as reflecting a freemium service model—a very familiar and common system in Silicon Valley where basic software is initially offered for free to capture market share and build a strong user community before charging fees for heavy commercial and enterprise use that consumes substantial resources.
Developer platforms and artificial intelligence agents
Concurrent with the major rollout of the new model, Alibaba launched the Qianwen Open Platform last Monday, an important strategic step aimed at providing unprecedented access for third-party developers and independent entities. The platform allows innovators and programmers to build custom artificial intelligence agents that operate efficiently and compatibly across a variety of smart devices, including mobile phones, personal computers, and wearable smart glasses. The initiative aims to create an integrated and vibrant technical environment enabling the development of practical applications primarily relying on the model’s advanced capabilities. Additionally, the company launched QwenWork, a dedicated workspace environment for enterprise AI agents that entered public beta testing the same day via web apps and desktop software, helping boost productivity and administrative efficiency. These sequential rollouts are part of the company’s strategy to build an integrated ecosystem for future technologies ahead of its anticipated unaudited second-quarter financial results announcement before the U.S. market opens on August 20.
Frequently asked questions
Question: What is the size of Alibaba Group’s new artificial intelligence model?
Answer: Named Qwen 3.8 Max, the model includes approximately 2.4 trillion parameters, but relies on an architecture that activates only 95 billion parameters in real time to reduce costs and increase operational efficiency.
Question: How does the company intend to generate financial returns from this open model?
Answer: The company plans to implement a new profit-sharing policy requiring major commercial users generating massive revenues to pay a percentage of their profits resulting from the use of the model’s open weights.
Question: What new platforms were launched for developers and enterprises?
Answer: The company launched the Qianwen Open Platform to develop smart agents for various devices, alongside QwenWork, designed to support enterprise work environments using advanced AI technologies.
