HANGZHOU, China: Alibaba unveils Qwen 3.8, a 2.4 trillion-parameter model its team ranks “second only to Fable 5” among all AI systems.
The July 19 announcement landed during the World AI Conference in Shanghai. A preview called Qwen3.8-Max-Preview went live the same day through Alibaba’s Token Plan, Qoder, and QoderWork at 10% of standard pricing. Investors liked what they heard: Alibaba shares rose as much as 5.4% Monday morning.
The timing is the story. Qwen 3.8 arrived three days after Moonshot AI’s Kimi K3, the 2.8 trillion-parameter model that briefly held the heavyweight crown alone. Alibaba owns roughly 36% of Moonshot, which makes this an in-house rivalry as much as a market one. It is also the fourth frontier launch in eleven days, after Grok 4.5 on July 8 and GPT-5.6 on July 9.
Some claims deserve scrutiny. The 2.4 trillion figure is Alibaba’s own, with no model card published. The “second only to Fable 5” ranking rests on internal evaluation, with no benchmark table released. The active-parameter count, the number that actually determines serving cost in a sparse model, stays undisclosed.
What is confirmed still matters. Developer Shuai Bai says this is Qwen’s first multimodal model above a trillion parameters, processing images, videos, and documents. The API speaks both OpenAI and Anthropic protocols, so existing coding agents can point at it without rebuilding anything.
Open weights are promised “soon,” with no date or license. Worth noting: Alibaba’s last two Max flagships were promised similar treatment and shipped closed.
Predecessor Qwen 3.7-Max scored 80.4% on SWE-bench Verified and led Terminal-Bench 2.0. If 3.8 clears those marks when independent numbers arrive, the frontier officially has a fourth Chinese contender.
Previous
