Alibaba Previews Qwen3.8-Max: 2.4 Trillion Parameters, 'Second Only to Fable 5' — But No Benchmark Table, Model Card, or License

MarkTechPost (corroborated by The Decoder, Dataconomy, Hacker News)·high signal

Alibaba's Qwen team previewed Qwen3.8-Max-Preview on July 19 at WAIC Shanghai, a 2.4T-parameter multimodal model it claims ranks just below Anthropic's Fable 5 on internal evals — two days after Moonshot shipped the 2.8T open-weight Kimi K3. The preview is live via Token Plan, Qoder and QoderWork at 10% of standard pricing, but the benchmark table, model card and license are all absent, and Alibaba has not disclosed active parameter count at inference — the number that actually matters for a sparse MoE. An open-weight release is promised 'soon,' which would break Alibaba's pattern of keeping Max-tier models closed.