Alibaba launches Qwen3.8, a 2.4-trillion-parameter model it ranks second to Fable 5

Alibaba’s Qwen team launched Qwen3.8 on July 19, 2026, a 2.4-trillion-parameter model the company called one of the most powerful available today and ‘second only to Fable 5.’ It has published no third-party benchmarks to support that ranking.

The release lands days after Moonshot AI’s 2.8-trillion-parameter Kimi K3, part of a widening race among Chinese labs to ship ever-larger frontier models. Alibaba says Qwen3.8 is its first multimodal model to pass 1 trillion parameters, processing images, video and documents, and expects it to beat the earlier Qwen3.7-Max on coding and complex productivity work.

A preview version, Qwen3.8-Max-Preview, went live the same day on Alibaba Cloud’s Token Plan and the Qoder coding platform. Alibaba’s tiered Token Plan runs from a Lite tier to a Pro tier offering six to eight concurrent agents, which the company advertises at around 40% off standard pay-as-you-go pricing. Alibaba said the full model will go open-weight ‘soon,’ without giving a date.

Those figures are Alibaba’s own. The company has released no independent evaluations, and parameter count alone does not establish capability — a model’s training data, architecture and post-training work matter as much as raw size, and rivals have made similar frontier claims on similar scales.

The open-weight release is the test to watch. If the weights ship on a stated date, outside researchers can check the frontier claim against real evaluations; if ‘soon’ keeps slipping, the ranking stays a marketing line.