Overview
- Alibaba unveiled Qwen3.8‑Max, its largest model to date, which it revealed on August 3 and says has 2.4 trillion parameters in a mixture‑of‑experts design.
- The model’s MoE architecture activates roughly 95 billion parameters at inference to cut compute cost and latency and Alibaba reports strong internal results on coding and multimodal tasks.
- Alibaba told developers it will publish the model weights for public download next week, a move meant to drive developer traffic back to Alibaba Cloud for compute, storage and hosting.
- Market reaction was positive after the announcement, but analysts warn the open‑weight label does not guarantee unrestricted use because final license terms and export controls will shape commercial adoption.
- Qwen3.8‑Max puts Alibaba in direct competition with Moonshot’s Kimi K3 and Baidu’s ERNIE, and the key next steps are independent benchmark verification and whether free model access converts into sustained paid cloud revenue.