Microsoft has recently introduced two in-house developed AI models, MAI-Voice-1 and MAI-1-preview, representing a significant leap forward in its foundational large model sector, with the objective of decreasing its dependence on OpenAI's models. Historically, Microsoft's AI prowess has been heavily reliant on OpenAI, with Microsoft promptly incorporating each new model released by OpenAI into its Azure platform and product offerings. The newly launched MAI-Voice-1 is a voice synthesis model adept at generating high-quality, real-time voice output on a single GPU, making it ideal for applications such as news broadcasting and podcast dialogues. MAI-1-preview, on the other hand, is a foundational large model that employs a Mixture of Experts (MoE) architecture and has been trained using around 15,000 NVIDIA H100 cards. Designed as a versatile model for efficient instruction execution and daily Q&A interactions, it has been made available for public testing on LMArena and is slated for gradual integration into Copilot's text applications. Microsoft has announced its commitment to continue developing a diverse range of in-house models, fostering the integration of proprietary and open-source models, and delivering efficient AI capabilities to users globally.