2 Sources
2 Sources
[1]
Alibaba releases trillion-parameter AI model to rival OpenAI, Google
The Hangzhou-based company released this new model from the Qwen3 series earlier in May this year. The series initially consisted of models ranging from 600 million to 235 billion parameters. Parameters are entities that decide how accurate an AI model is. However, the more parameters, the more energy and computing power are required to tune them and keep the system running. OpenAI's GPT-4.5 is known to be one of the world's biggest AI models, with an estimated parameter count of 5 to 7 trillion. The Qwen-3-Max-Preview model is a text-only model that outsmarted its best model, the Qwen3-235B-A22B-2507 released in July. Alibaba also released benchmark results indicating that Qwen-3-Max-Preview outperformed MoonShot AI's Kimi K2, a non-reasoning version of Anthropic's Claude Opus 4, and DeepSeek V3.1 across five tests. However, an official technical report did not back the findings. "Qwen3-Max-Preview shows substantial gains ... in overall capability, with significant enhancements in Chinese-English text understanding, complex instruction following, handling of subjective open-ended tasks, multilingual ability, and tool invocation. Scaling works - and the official release will surprise you even more," the company said.
[2]
Alibaba releases Qwen-3-Max-Preview, its largest AI model yet
Alibaba has released Qwen-3-Max-Preview, its most powerful artificial intelligence model so far. The model has over one trillion parameters, making it the largest system the company has built to date. The model is text-only and was made available on September 5, 2025, through Alibaba Cloud and the OpenRouter platform. Unlike some AI models, it is not open-source. Access is limited to official channels, which helps control how the model is used. According to Alibaba, Qwen-3-Max-Preview performs better than previous versions. Internal tests suggest it also outperforms many other models, though these results have not been verified independently. Users can access Qwen-3-Max-Preview through Alibaba's cloud services. The pricing is structured as follows: Model Studio activation is required to use the system, ensuring proper use and management of resources. Alibaba plans to invest $52 billion in AI infrastructure, research, and development over the next three years. This funding will support the development of new models, cloud services, and AI tools. The release of Qwen-3-Max-Preview and the investment plan show that Alibaba is expanding its AI capabilities and building infrastructure to support large-scale AI applications.
Share
Share
Copy Link
Alibaba has released its largest AI model to date, Qwen-3-Max-Preview, boasting over one trillion parameters. This text-only model aims to compete with industry giants like OpenAI and Google, showcasing significant improvements in various capabilities.
In a significant move that positions it alongside tech giants like OpenAI and Google, Alibaba has unveiled its most powerful artificial intelligence model to date: Qwen-3-Max-Preview. This text-only model, released on September 5, 2025, boasts over one trillion parameters, marking a substantial leap in Alibaba's AI capabilities
1
2
.The Qwen-3-Max-Preview is part of Alibaba's Qwen3 series, which initially ranged from 600 million to 235 billion parameters. Parameters are crucial entities that determine an AI model's accuracy, with higher numbers generally indicating more sophisticated capabilities
1
. Alibaba claims that this new model outperforms its previous best, the Qwen3-235B-A22B-2507, released in July1
.According to Alibaba, the Qwen-3-Max-Preview demonstrates "substantial gains ... in overall capability, with significant enhancements in Chinese-English text understanding, complex instruction following, handling of subjective open-ended tasks, multilingual ability, and tool invocation"
1
. The company also reported that their internal benchmarks show the model outperforming competitors such as MoonShot AI's Kimi K2, a non-reasoning version of Anthropic's Claude Opus 4, and DeepSeek V3.1 across five tests1
.Source: Interesting Engineering
Unlike some AI models in the market, Qwen-3-Max-Preview is not open-source. Access to the model is restricted to official channels, including Alibaba Cloud and the OpenRouter platform
2
. This controlled access helps Alibaba manage how the model is used and ensures proper resource allocation. Users interested in leveraging the model's capabilities must activate Model Studio, a step designed to ensure appropriate use and resource management2
.The release of Qwen-3-Max-Preview is part of Alibaba's broader strategy to expand its AI capabilities. The company has announced plans to invest a staggering $52 billion in AI infrastructure, research, and development over the next three years
2
. This substantial investment underscores Alibaba's commitment to building robust infrastructure capable of supporting large-scale AI applications and developing cutting-edge AI tools and cloud services.Related Stories
Alibaba's trillion-parameter model enters a competitive landscape dominated by tech giants like OpenAI and Google. For context, OpenAI's GPT-4.5 is estimated to have between 5 to 7 trillion parameters, setting a high bar in the industry
1
. While Alibaba's internal tests suggest strong performance, it's worth noting that these results have not been independently verified, and an official technical report is yet to be released1
.As AI models grow in size and complexity, they face increasing challenges related to energy consumption and computing power. The trillion-parameter scale of Qwen-3-Max-Preview implies significant resource requirements for tuning and operation
1
. This aspect raises important questions about the sustainability and scalability of such large AI models in the long term.Summarized by
Navi
[1]
27 Mar 2025•Technology
06 Mar 2025•Technology
28 Jan 2025•Technology
1
Business and Economy
2
Technology
3
Business and Economy