• 2 min read
Alibaba unveils 2.4T-parameter Qwen3.8 with open weights pledge
Alibaba says its new Qwen3.8 has 2.4 trillion parameters, multimodal support, and open weights coming soon.
Image: ITzine
Alibaba has introduced Qwen3.8, a new flagship language model the company says has 2.4 trillion parameters, multimodal support, and benchmark results that, in its own testing, trail only Anthropic’s Claude Fable 5. The model was shown publicly at the World Artificial Intelligence Conference in Shanghai and has already been rolled out to some users through the Alibaba Token Plan, as well as via Qoder and QoderWork.
The size jump is significant. A year ago, Alibaba’s flagship Qwen3-Max had roughly 1 trillion parameters, while the open Qwen3.5 released in February came in at 397 billion. Against that backdrop, Qwen3.8 looks less like a marketing flourish and more like a substantial engineering step.
More notable than the parameter count, though, is Alibaba’s promise to release the model’s weights. The Qwen team said on X that Qwen3.8's weights will be published “in the near future.” If that happens, it would mark a return to a format developers can inspect and run beyond a cloud API. In recent months, Alibaba has kept the most powerful Max models closed, offering access only through the cloud. Qwen3.7-Max, released in May, never received open weights despite demand from developers.
The announcement also lands in a market increasingly focused on sheer scale as well as quality. Just days before Alibaba’s reveal, Moonshot AI introduced Kimi K3 with 2.8 trillion parameters, raising the bar again in China’s model race.
Globally, OpenAI, Anthropic, and Google still set the pace on quality and capabilities. Chinese developers are responding with models aimed at matching Western flagships not just in features, but in size and breadth. Independent testing will determine how close Qwen3.8 really is to the leaders. For now, Alibaba has delivered one of the loudest signals yet in China’s AI model race—and left one crucial question hanging: whether its new flagship will remain a closed service or become a tool developers can truly work with.
Recommended reading
Anthropic adds Fable 5, then halves usage limits
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via ITzine
