Alibaba used the World Artificial Intelligence Conference in Shanghai to preview its most ambitious model yet. On July 19, the company unveiled Qwen3.8-Max-Preview, a 2.4-trillion-parameter system it calls one of the most powerful available today — "second only to Fable 5," Anthropic's frontier model.
A jump in scale
Going from roughly 1.8 trillion to 2.4 trillion parameters in a matter of months puts Qwen in the same bracket as Moonshot AI's Kimi K3 (2.8T), the largest open model to date. It is also the Qwen team's first model above one trillion parameters with native multimodality, able to handle images, video and documents rather than text alone.
Closed today, open 'soon'
The important caveat is availability. Despite framing that leans on the open-source label, Qwen3.8-Max debuted as a proprietary preview, reachable through Alibaba's API "Token Plan," Qoder and QoderWork. Alibaba says open weights "will follow" — but its last three Max-tier flagships shipped closed and never received open weights, so the open release remains a promise rather than a fact.
An unverified ranking
The "second only to Fable 5" claim carries a second asterisk: it rests entirely on Alibaba's own internal evaluations. No third-party benchmark results have been published, and the company offered comparative positioning rather than public scores. On the raw-size leaderboard, Kimi K3 remains larger, so this is not China's biggest model either.
The WAIC backdrop
The preview landed amid a torrent of Chinese model announcements at WAIC, where domestic labs are racing to close the gap with US frontier systems. Naming Anthropic's model as the bar to beat is itself a signal — Chinese developers increasingly measure themselves against a specific Western frontier rather than the field in general.