Alibaba has released the open weights for Qwen3.8-2.4T-A95B (also known as Qwen3.8-Max), its largest open-weight model, designed to bring near-frontier capabilities to the open ecosystem.
- The model features 2.4 trillion total parameters with 95 billion activated per token.
- It utilizes a fine-grained mixture of experts (MoE) architecture combined with a hybrid of full and linear attention mechanisms.
- The release highlights serving this large-scale model on NVIDIA GB300 NVL72 hardware.