Inside Alibaba's Two New Models, Qwen3.8-Max and Qwen3.8-27B

Alibaba released downloadable weights for Qwen3.8-Max, a 2.4 trillion-parameter mixture-of-experts vision-language model, alongside the smaller Qwen3.8-27B, within a week of unveiling the Max model on August 2. This marks the first time Alibaba has open-sourced weights for its top-tier Max line, though the released weights are text-only and lack the full 1-million-token context window of the hosted API version. Qwen3.8-Max ranks fifth overall and second among open models on Artificial Analysis' Intelligence Index, tops the 𝜏³-Banking benchmark, and places highly on vision and web-dev coding arenas, though it generates more tokens per task, raising effective cost. Qwen3.8-27B, released under Apache 2.0, is described as the most capable model that can currently run locally on consumer hardware. The release follows competitive pressure from Moonshot AI's Kimi K3 and Z.ai's GLM-5.3 in the open-weights race among Chinese AI labs.