Alibaba's 2.4T Qwen3.8 Claims No.2 Spot, Behind Only Fable 5

Alibaba's Qwen team dropped Qwen3.8-Max-Preview on July 19, and the pitch is unusually blunt: 2.4 trillion parameters, and "second only to Fable 5." That's a Chinese lab publicly ranking itself against a specific frontier model rather than a leaderboard average — and it's the team's first multimodal model past the trillion-parameter line, taking text, images, video, and documents.

The catch is what's missing. No benchmark tables, no technical report, no independent evals. It's a sparse Mixture-of-Experts model, and Alibaba hasn't disclosed active parameters per token — the one number that determines what serving it actually costs. "2.4T" is a headline figure, not a serving figure.

Access runs through Alibaba's Token Plan, Qoder, and QoderWork at 10% of standard pricing — $6/week for 2,500 credits on Lite, $68/week for 40,000 on Pro with six to eight concurrent agents. There's no standalone per-token API price yet. Open weights are promised "soon," but with no date, no license, and no Hugging Face repo — and the timing lands days after Moonshot shipped Kimi K3 with weights actually in hand.

Why it matters: a 2.4T open-weight multimodal model would reset the frontier for self-hosting — if the weights and the benchmarks ever show up.