★ Featured Guide · Guide
Alibaba previewed Qwen3.8-Max on July 19, 2026 at the World AI Conference in Shanghai — 15 days before this article. The current accessible endpoint is qwen3.8-max-preview, not a finished release. Confirmed facts: 2.4 trillion parameter sparse MoE, multimodal (text, image, video, documents), 1M context window (from Qwen Code integration notes), always-on reasoning. Alibaba claims it is "second only to Fable 5." What is still missing: no benchmark table, no model card, no published per-token API pricing, no active parameter count, no Hugging Face weights, no open-weight date or license. The $2/$6/M pricing circulating on X is not confirmed.
Aug 3, 2026
·
6 min read