Alibaba previews 2.4T-parameter Qwen model without benchmarks

Alibaba Group Holding Ltd. has previewed Qwen3.8, a 2.4 trillion-parameter AI model that the company says ranks behind only Anthropic PBC’s Claude Fable 5 among frontier models. The announcement pushes Alibaba’s Qwen line into the multi-trillion-parameter tier, but the company has not published the benchmark data needed to evaluate that ranking.

The model was introduced at the World Artificial Intelligence Conference in Shanghai as Qwen3.8-Max-Preview. Alibaba says it is the first Qwen model with more than 1 trillion parameters that can handle images, video and documents in addition to text.

Alibaba’s Qwen team said in a post on X that Qwen3.8 is “second only to Fable 5.” That is Alibaba’s assertion. No independent evaluation has confirmed the placement, and the company did not release benchmark scores with the preview.

What Alibaba did not publish

The missing data is material because Alibaba has previously provided fuller disclosures for major Qwen releases. Qwen3.7-Max, released in May, came with published results, including a 56.6 score on the Artificial Analysis Intelligence Index.

For Qwen3.8-Max-Preview, Alibaba has not released a model card, an activated-parameter count or any benchmark package. The company also has not provided task-level comparisons against Qwen3.7-Max, making it difficult to tell where the new system improves beyond scale and multimodal support.

Alibaba described Qwen3.8 as “continuously evolving” and said open weights would arrive “soon.” It did not give a release date for those weights or disclose license terms. Until those details are available, developers and enterprise buyers have only access to the preview build, not the open-weight artifact Alibaba has said is coming.

The preview is available through Alibaba’s Token Plan subscription and through its Qoder and QoderWork developer platforms. Alibaba is pricing access at 10% of the standard rate during the trial period.

China’s trillion-parameter race

The Qwen3.8 preview landed three days after Beijing Moonshot AI Technology Co. Ltd. released Kimi K3, a 2.8 trillion-parameter model. Together, the launches show Chinese AI developers pushing into a scale bracket that had been associated mainly with the largest U.S. labs.

Alibaba has made Qwen a central part of its bid to become a default AI supplier in China. Over the past year, the company has released a series of open-weight models while expanding cloud infrastructure and custom chips designed to run them. That strategy has helped Qwen build one of the larger developer communities among Chinese model families.

The enterprise read is straightforward: Alibaba is signaling that Qwen is moving upmarket, but the evidence is incomplete. The company’s comparison with Anthropic’s Claude Fable 5 is not backed by public scores, and the last ranked Qwen flagship, Qwen3.7-Max, sits below Fable 5 on LMArena, where Fable 5 is ranked No. 1.

Benchmarks, a model card and the promised open weights would give buyers and developers a basis for comparison. Alibaba has not released those items yet.

This story draws on original reporting from SiliconANGLE.