Alibaba previews 2.4-trillion-parameter Qwen 3.8-Max at WAIC

Qwen 3.8-Max, previewed at the World AI Conference in Shanghai, is Alibaba's new flagship: a 2.4-trillion-parameter model built on a sparse Mixture-of-Experts architecture, confirmed multimodal across text and visual inputs with reported support for video, documents, speech and image generation. Alibaba claims performance 'second only to Fable 5,' but public benchmark tables have not been released, so the ranking remains a company assertion for now.

Availability is currently preview-only, through Alibaba's Token Plan subscription and Qoder/QoderWork platforms. To build momentum, Alibaba's Qwen team launched the #QwenGrowthPlan, inviting developers to complete real-world tasks with Qwen 3.8 and submit results — an explicit push to demonstrate agentic capability and gather feedback.

Strategically, Qwen 3.8-Max reinforces the week's dominant subplot: the rise of capable Chinese models. Reporting that Chinese AI models now handle nearly a third of enterprise tokens at roughly one-tenth the cost of U.S. rivals, combined with DeepSeek's V4 and Moonshot's Kimi K3 (now on Azure and Perplexity), underlies the 'US AI bubble' anxiety visible across Reddit this week. A 2.4T-parameter multimodal Qwen adds to the sense that the capability gap is narrowing fast.

Caveats: 'second only to Fable 5' with no published benchmarks is unverifiable, and parameter count is a weak capability proxy — the whole week's competitive story was about efficiency and cost, not raw size. Preview-only access and platform lock-in (Token Plan, Qoder) also limit near-term impact. What to watch is the benchmark release and whether Qwen 3.8-Max ships open weights, which would sharpen its competitive threat considerably.