Alibaba Qwen 3 8 Image © Alibaba Alibaba Qwen 3 8 (Image © Alibaba )

Currently, Qwen-3.8 is available exclusively as a preview version via Alibaba’s cloud infrastructure. Access is provided through a token-based model, which means that users are limited to a specific token budget and do not have unlimited API access. The “Max” variant of the model is the main version being offered during this preview phase.

Release of the Sail-AI Software Stack

As part of the World AI Conference, T-Head, Alibaba’s chip development division, has begun releasing the Sail-AI software stack. This release currently includes six repositories on GitHub, including the “Triton” inference framework as well as software adapted from Nvidia and Deepseek

The developer documentation indicates that these public repositories represent only a portion of the complete Sail software suite. The strategic goal behind releasing this code is to lower the barriers for developers who wish to transition to Alibaba’s proprietary Zhenwu accelerators. These hardware accelerators are used internally by Alibaba to run its models, including Qwen-3.8, and have already been shipped to a significant number of external customers.