Microsoft Completes Pre-Training of 2.8-Trillion-Parameter MAI-Base-2

Microsoft’s AI division has completed pre-training MAI-Base-2, a large-scale model built with a Mixture-of-Experts (MoE) architecture. It has 2.8 trillion total parameters, with 80 billion active for any given task.

Microsoft says MAI-Base-2 is 4 to 12 times more efficient to pre-train than DeepSeek-V4-Pro-Base. The company trained the model for 38 days using 16,000 Nvidia GB200 GPUs.

Unlike its predecessor, MAI-Base-2 was pre-trained with vision capabilities built in. It is undergoing post-training and is not yet publicly available. Microsoft plans to license the model to partners through its Foundry platform.