Alibaba has launched Qwen3.8-Max, the latest flagship model in its Qwen family, describing it as its most capable AI system to date. The model is now available through Alibaba Cloud’s Model Studio APIs and the company’s QwenWork platform, with open weights scheduled for release next week.
Built on the Qwen 3.5 architecture, Qwen3.8-Max contains 2.4 trillion parameters, although its sparse mixture-of-experts architecture activates only 95 billion parameters during inference through a sparse mixture-of-experts design. Alibaba says this architecture reduces computational costs and latency compared with dense models of a similar size.
With 2.4 trillion parameters, Qwen3.8-Max is one of the largest AI models released to date. It is slightly smaller by total parameter count than Moonshot AI’s 2.8 trillion-parameter Kimi K3, although parameter counts alone do not determine model performance. Alibaba says Qwen3.8-Max supports multimodal input and a context window of up to 1 million tokens.
A push toward AI agents, not just chatbots
Alibaba is positioning Qwen3.8-Max around long-running tasks where AI systems are expected to plan, execute and refine work with limited human involvement.
The company highlighted internal tests where the model independently developed a software engineering project over 16 days, using feedback loops, testing and analysis to improve the result. Alibaba said the model also reproduced and improved a research paper’s experiments, competed in an online challenge against hundreds of human teams, and completed other multi-step tasks.
For businesses, the focus is shifting from asking AI questions to using AI as a digital worker. Alibaba said Qwen3.8-Max can handle tasks such as reviewing large document collections, creating software applications from screenshots, analyzing videos, generating designs and supporting professional workflows.
Open weights could help Alibaba reach more developers
One of the biggest changes with Qwen3.8-Max is Alibaba’s decision to release open weights for a Max-series model for the first time.
Open-weight models allow developers to download the underlying model files, customize them and run them independently, although a model of this size still requires significant computing infrastructure. The move reflects a broader strategy among Chinese AI companies to attract developers by offering access to powerful models rather than keeping them entirely behind closed platforms.
The planned release differs from the strategy used for the flagship proprietary systems offered by OpenAI, Anthropic, and Google, which generally do not disclose parameter counts or make their most advanced model weights downloadable.
Bigger models bring bigger challenges
Qwen3.8-Max also shows the limits of today’s AI race. A model with trillions of parameters still requires significant computing resources, meaning many organizations will not be able to run it independently.
There are also questions around reliability. AI companies increasingly highlight autonomous demonstrations, but long-running tasks can still fail because of incorrect reasoning, unexpected software errors or poor decisions.
Alibaba’s challenge will be proving that Qwen3.8-Max can move beyond impressive demonstrations and deliver consistent results for everyday users and enterprises. With open weights arriving soon, developers will have the chance to test whether Alibaba’s latest model can compete outside company-controlled benchmarks and become a serious alternative in the global AI market.
Also read: Enterprise AI Copilots Give Way to Autonomous Workflows to learn how businesses are moving from AI assistants toward systems that can independently execute complex tasks.