Alibaba launches Qwen3.8-Max AI model with 2.4 trillion parameters

Alibaba announced the Qwen3.8-Max model, a 2.4 trillion‑parameter AI system featuring a sparse‑Mixture‑of‑Experts architecture that activates about 95 billion parameters at a time. The model supports a roughly 1 million‑token context window and is built for autonomous, long‑horizon coding agents that can run multi‑step projects for up to 16 days without human intervention.

Benchmarks reported by Alibaba show the model scoring 86.1 on the OSWorld‑Verified benchmark, ranking fourth on the Frontend Code leaderboard, second in Vision Arena, and fifth in Text Arena. It achieved 67.7 on SWE‑bench Pro and strong results on Terminal‑Bench and PaperBench. Pricing is positioned below most frontier models, and Alibaba plans to release the weights under a yet‑to‑be‑confirmed license. The release raises governance questions under the EU AI Act as enterprises consider integrating the model into workflows.