Alibaba Releases Its Most Powerful AI Model to Date, Which Is Slightly Smaller Than Moonshot’s

China’s Alibaba (9988.HK) opened a new tab on Monday and revealed what it claimed to be its largest and most capable artificial intelligence model, the Qwen3.8-Max, which is comparable in size to a product from domestic competitor Moonshot AI that was introduced last month.

Chinese IT firms, a major player in open-weight AI models worldwide, are engaged in a quick and intense struggle to develop more potent systems without making them unaffordable to operate.

Qwen3.8-Max has 2.4 trillion parameters, which are numerical settings that a model learns from data and utilizes to identify patterns, provide solutions, and perform tasks. The Kimi K3 from Moonshot contains 2.8 trillion parameters.

Although a greater number does not always indicate a better model, it has become a frequently monitored indicator of the size of the data and processing power underpinning sophisticated AI systems.

In order to help their models find momentum among developers, Chinese tech companies are eager to reveal parameter counts. Their models are often open-weight, which means that developers can run or modify the system by downloading the underlying learned settings.

In contrast, OpenAI, Anthropic, and Google (GOOGL.O) do not release the number of parameters for their closed-source models.

Qwen3.8-Max was released on the crowdsourced model comparison website Arena.AI, where it quickly rose to the top of the Chinese text model rankings, although it currently behind three Anthropic Opus variations and Claude Fable 5.On Arena, though.

Qwen3.8-Max came in second place worldwide on AI’s leaderboard for AI models that analyze photos and other visual content, just surpassed by a Claude Fable 5 variation.

Both Kimi K3 and Qwen3.8-Max can process up to one million tokens at once and handle text, pictures, and video.

A significant figure indicates that the model can process enormous volumes of data at once, such as lengthy legal files, a sizable software codebase, or hundreds of pages of documents. Tokens are bits of data, frequently fragments of words or short sentences.