Alibaba drops Qwen 3.8, says it trails only Fable 5

Alibaba just threw a 2.4-trillion-parameter wrench into the open-weight race. The company unveiled Qwen 3.8, a multimodal model it claims is “second only to Fable 5” in performance — a remarkably specific (and confident) boast for a release with zero public benchmarks to back it up yet.

Developer Shuai Bai says this is the team’s first multimodal model to cross the trillion-parameter threshold, capable of handling images, video, and documents alongside text. That’s a serious expansion in capability. The preview is available now through Alibaba’s Token Plan, Qoder, and QoderWork platforms at just 10 percent of the standard price. Open weights are promised “soon,” though Alibaba hasn’t specified a date.

The real intrigue isn’t just the model’s size — it’s the timing. Moonshot AI’s Kimi K3 has been building momentum with a similar playbook: dangle the promise of open weights while funneling users into a proprietary chat app and API. Alibaba’s release threatens to short-circuit that strategy entirely. Why wait for Moonshot to deliver when Qwen gives you comparable (or better) capabilities right now, with actual open weights on the near horizon?

Moonshot isn’t exactly hurting — Bloomberg reports the company hit $300 million in annual recurring revenue by June and is eyeing a public listing within six months. But a well-timed open-weight release from a giant like Alibaba can cool investor enthusiasm fast. For developers, this is the kind of competition that actually matters: two serious players racing to put frontier-ish capabilities in your hands without the walled garden. The question isn’t whether Qwen 3.8 is good. It’s whether Moonshot can keep its funnel intact long enough to ring the bell on Wall Street.