Microsoft Unveils In-House AI Model MAI-Base-2, Touting Significant Efficiency Gains

Microsoft has completed pre-training its proprietary AI model, MAI-Base-2, which features 2.8 trillion total parameters and 80 billion active parameters. Microsoft AI CEO Mustafa Suleyman announced the model outperforms OpenAI's model powering GPT-5.6 Sol while using only one-quarter of the compute power. The model was trained over 38 days using 16,000 Nvidia GB200 GPUs across roughly 50 trillion tokens, delivering a 4.2 to 12.3-fold training efficiency advantage over DeepSeek-V4-Pro-Base. Microsoft plans to license the model to enterprise clients through Azure AI Foundry as it works to reduce reliance on third-party developers.