Alibaba's Qwen3.8 packs 2.4 trillion parameters — and it's already available in preview

Alibaba’s Qwen team dropped a quiet bombshell on Sunday. Qwen3.8 — a 2.4 trillion parameter model — is coming soon, and it’s claiming a spot right behind Anthropic’s Fable 5 in the global AI rankings. The preview version is already live.

Qwen3.8-Max-Preview is available now on Alibaba’s Token Plan platform, Qoder, and QoderWork. The full open-source release will follow, according to the Qwen team, who described the model as evolving “by the day.”

The jump from Qwen3.7-Max to 3.8 is substantial. Internal benchmarks show the preview model matching frontier-level performance on complex, multi-agent, long-horizon tasks — full-stack software development, data analysis, and Office productivity workflows. These are exactly the kinds of real-world evaluations that matter more than static leaderboard scores.

What makes the 2.4 trillion parameter count notable is context. That puts Qwen3.8 in the same weight class as the largest models ever built. The team’s own characterization — “possibly the most powerful model besides Fable 5” — is a direct challenge to the current hierarchy, where Anthropic’s flagship has held the top spot on several benchmarks since its release.

Alibaba hasn’t shared the full technical report yet. No architecture details, no training compute figures, no specific benchmark numbers. But the decision to push a preview build to production before the full paper suggests the team is confident enough to let developers start kicking the tires right now.