Microsoft's research lab, Microsoft AI, has released three new foundational AI models, MAI-Transcribe-1, MA-Image-2, and MA-Declan, which can generate text, voice, and images. The release signals Microsoft's continued efforts to develop a larger stack of multimodal AI models. Despite this, Microsoft remains tied to OpenAI. The models were developed by Microsoft's MAI Superintelligence team, led by Mustafa Suleyman, and are being released on Microsoft Foundry and available in MAI Playground. The company hopes these models are cheaper than those from Google and OpenAI, and will be featured in Microsoft products and experiences.