Alibaba's shares rose more than 6% in Hong Kong on Monday following the unveiling of Qwen3.8-Max, which the company describes as its most powerful artificial intelligence model to date. The release intensifies competition among Chinese AI developers and challenges Western counterparts.

The new model features 2.4 trillion parameters, placing it just behind Moonshot AI's Kimi K3, which launched last month with 2.8 trillion parameters. Parameter counts have become a key metric in the AI race, though they don't guarantee superior performance.

Read also
Stocks
AstraZeneca drops 6% on BMS merger talks; investors question logic
AstraZeneca shares tumbled more than 6% on reports of preliminary merger discussions with Bristol Myers Squibb, a deal that would create a nearly $400 billion pharma giant but drew skepticism from investors.

Parameter race and open-weight strategy

Chinese AI firms increasingly highlight parameter counts as they compete for developer and enterprise adoption. Unlike OpenAI, Anthropic, and Google, which keep such details private for closed-source models, Chinese companies often publish these figures because many of their models are open-weight, allowing developers to download and modify them.

Strong global rankings

On Arena.AI, a crowdsourced model comparison platform, Qwen3.8-Max immediately became the highest-ranked Chinese model for text generation. However, it still trails Anthropic's Claude Fable 5 and three Claude Opus variants on the overall text leaderboard. For multimodal tasks involving images and visual content, it ranks second globally, behind only one Claude Fable 5 variant.

Like Kimi K3, Qwen3.8-Max can process text, images, and video while supporting context windows of up to one million tokens. This allows the model to analyze extensive datasets in a single request, such as lengthy legal documents or large codebases.

Mixture-of-experts design cuts costs

Alibaba said Qwen3.8-Max uses a mixture-of-experts architecture, activating only about 95 billion of its 2.4 trillion parameters per request. This approach reduces computing costs and improves response speed, making the model more practical for commercial deployment.

Benchmark tests show the model is broadly competitive with leading AI systems from OpenAI and Anthropic. It outperformed several US rivals on coding, engineering, and multimodal benchmarks, though it trailed on some general-purpose reasoning tasks.

AI competition heats up

The launch underscores Alibaba's growing commitment to AI as it competes with domestic rivals like DeepSeek, Baidu, Tencent, and Moonshot AI. China has become a major source of open-weight AI models, with developers racing to improve performance while keeping inference costs low.

Alibaba has steadily expanded investment in AI infrastructure and cloud computing. In May, the company said it would exceed its previously announced plan to invest up to 380 billion yuan ($55.96 billion) in AI infrastructure over three years. Qwen3.8-Max will be available on Alibaba Cloud's Model Studio platform next week, targeting enterprise developers.

The model also completed a software engineering project in 16 days, highlighting its coding capabilities. This release is expected to further intensify competition in China's AI market as companies continue investing heavily in large language models.

This article is for informational purposes only and does not constitute financial advice.