Qwen3.8-Max can process visual information as well as text and has been designed for coding, research, workplace applications, and tasks that require extended autonomous operation. (Image: Qwen.AI)China’s AI companies appear to be narrowing the gap with their US rivals. Chinese technology giant Alibaba on Monday, August 3, unveiled its latest AI model, Qwen3.8-Max. The launch came on the same day that Chinese AI research firm DeepSeek introduced its V4-Flash model, underscoring the rapid pace of AI development in China and the growing popularity of open-weight AI models.Alibaba has described Qwen3.8-Max as its largest and most capable AI model in its Qwen series to date. The new model comes with multimodal capabilities and has 2.4 trillion total parameters with a context window of up to 1 million tokens. Based on its total parameters, Qwen3.8-Max is behind its domestic rival, the recently launched Moonshot AI’s Kimi K3, which boasts 2.8 trillion parameters. Even though parameter count has become a widely recognised measure of the scale of the computing and data behind AI systems, the higher parameter count does not automatically translate to better model on its own.The new model is capable of processing visual information as well as text and has been designed for coding, research, workplace applications, and tasks that require extended autonomous operation. The model was introduced on the crowdsourced, model-comparison platform Arena AI, and it soon became the highest-ranking Chinese model in terms of text. However, it still lags behind Claude Fable 5 and three variants of Opus from Anthropic. On Arena AI’s leaderboard for AI models that analyse images and other visual material, Qwen3.8-Max ranked second globally, only behind one of the Claude Fable 5 variants. Also Read | DeepSeek launches V4 Flash, one of the cheapest models to run: Is AI now a commodity?When it comes to model build, Qwen3.8-Max is based on a sparse mixture-of-experts architecture. According to Alibaba, the model completed a software-engineering project in a mere 16 days. Since it uses the mixture-of-experts design, it is able to divide work among specialised parts of the system instead of switching on the entire model for every request. The makers claim that only 95 billion parameters are used at a time, and this reduces costs considerably along with response delays. Qwen3.8-Max, just like Kimi K3, is capable of handling text, images, and video and is able to process up to one million tokens at a time. The model ranked fifth on the Text Arena model leaderboard, second on the Vision Arena, and fourth on the coding-focused arena. Alibaba said that the model is capable of analysing lengthy documents, television series, and livestreams and also turning content into searchable knowledge bases. The model is also capable of editing video footage, generating educational animations, building websites from screenshots, converting 2D floor plans into 3D designs, and even creating interactive games from simple prompts in natural language. When it comes to availability, the Qwen3.8-Max is available to international developers through application programming interfaces (APIs) on Alibaba Cloud’s Model Studio Platform. It can also be accessed via QwenWork, the company’s workplace AI platform. The Chinese tech giant has said that it plans to release the model’s weight next week, enabling developers to download and deploy it independently. Alibaba has priced Qwen3.8-Max at $2 per million input tokens and $6 per million output tokens, making it significantly cheaper than flagship models from OpenAI and Anthropic, which typically charge several times more for their most capable models.