Alibaba to Release Qwen3.8-Max Open-Weights Model Next Wednesday
Alibaba is scheduled to release Qwen3.8-2.4T-A95B (also known as Qwen3.8-Max), a massive new open-weights model, next Wednesday. The model has already been listed on the Chinese ModelScope platform ahead of its official launch. As Qwen is one of the leading open-weights LLM families, the release of a high-capacity model like Qwen3.8-Max could significantly advance the capabilities available to the open-source AI community. It offers a powerful alternative to proprietary frontier models for developers and researchers worldwide. The model's naming convention, Qwen3.8-2.4T-A95B, suggests a Mixture of Experts (MoE) architecture featuring approximately 2.4 trillion total parameters with 95 billion active parameters. This architecture allows the model to deliver high performance while optimizing computational efficiency during inference.
## BACKGROUND
Open-weight models are AI models whose trained parameters are publicly shared, allowing anyone to download and run them locally. Mixture of Experts (MoE) is an architectural design that routes inputs to specialized sub-networks ("experts"), enabling models to scale up parameter counts without a proportional increase in computing costs.