Chinese tech firm Alibaba launches powerful new AI model

Chinese technology firm Alibaba recently introduced Qwen3.8-Max, an advanced artificial intelligence model boasting 2.4 trillion parameters. Launching simultaneously with DeepSeek’s V4-Flash, this release highlights an intense race among tech companies in China to match or surpass their American competitors. Global developers are increasingly shifting their focus toward highly capable open-weight systems to democratize enterprise automation and reduce reliance on proprietary tech leaders.

Computing Power and Processing Scale

This newly released system natively processes both visual media and text, supporting an extensive context window of up to one million tokens. While its 2.4 trillion parameter size falls slightly behind Moonshot AI’s Kimi K3, which utilizes 2.8 trillion parameters, experts recognize that sheer parameter volume does not guarantee superior results. Qwen3.8-Max relies on a specialized sparse mixture-of-experts structure. Rather than running the entire network for every query, this targeted design activates only 95 billion parameters simultaneously during operations. This targeted routing method significantly lowers computing costs while also minimizing annoying response delays for users.

Performance Benchmarks and Enterprise Utility

Designed specifically for prolonged autonomous operations, the model recently executed a complex software engineering project entirely on its own over 16 days. It has also been rigorously tested on the crowdsourced evaluation platform Arena AI. Qwen3.8-Max secured the second global spot in analyzing images and visual material, trailing only one specific variant of Claude Fable 5. Furthermore, it achieved fifth place on the competitive text leaderboard and fourth in specialized coding evaluations. The system seamlessly analyzes lengthy documents, translates television livestreams, and edits video footage. It can even transform simple text prompts into fully functional interactive games, build websites from basic screenshots, and convert flat floor plans into three-dimensional architectural designs.

Global Pricing and Developer Access

International developers can immediately integrate this technology through application programming interfaces provided on Alibaba Cloud’s Model Studio Platform. It is also fully integrated into QwenWork, which serves as the company’s dedicated workplace productivity suite. Pricing sits at a highly competitive $2 per million input tokens and $6 per million output tokens. This aggressive cost structure makes it substantially more affordable than flagship models from market leaders like Anthropic and OpenAI. The tech giant plans to release the official model weights next week, empowering developers to download, modify, and deploy the system directly on their own private servers. ​​​​​​​