Alibaba claims Qwen3.8-Max is #2 globally, but lacks independent verification

Alibaba previewed Qwen3.8-Max on July 19, 2026, at the World AI Conference in Shanghai, making its boldest capability claim yet: a 2.4-trillion-parameter multimodal model that the company states ranks second only to Anthropic's Claude Fable 5. The model spans text, images, video, and documents, and Alibaba claims it outperforms its predecessor Qwen3.7-Max on coding, data analysis, and office workflows. However, no independent verification exists. No benchmark table, model card, license, or activated-parameter count accompanied the announcement. The model is absent from LMArena, Artificial Analysis, Hugging Face, and GitHub. Every capability claim traces back to Alibaba's own statements or early-access anecdotes.

The timing is significant: Qwen3.8-Max arrived two days after Moonshot AI released Kimi K3 (2.8 trillion parameters) with shipped open weights. Alibaba's promise of open-weight release uses vague language ('soon'), contrasting with Kimi K3's firm July 27 date. Additionally, naming confusion arises because 'Qwen3.8' is conflated with the unrelated 8.2-billion-parameter Qwen3-8B model from 2025. The preview is accessible via Alibaba's $6/month Token Plan subscription and its Qoder and QoderWork platforms, supporting tools like Claude Code and Cursor. Until independent benchmarks emerge, engineering and marketing teams should treat Alibaba's claims as unverified vendor statements.