Microsoft Launches In-House AI Models, Claims Up to 89% Cost Savings

Microsoft AI introduced two new internally developed AI models in public preview on Wednesday: MAI-Image-2.5-Pro, described as its highest-fidelity image generator, and MAI-Voice-2-Flash, a speech model designed for high-volume enterprise workloads.

The company stated these models offer significant cost reductions compared to those from OpenAI. The new models are already integrated into several Microsoft products, including Bing, PowerPoint, OneDrive, Dynamics 365, Excel, GitHub Copilot, and Azure, marking a shift towards relying on its in-house capabilities.

MAI-Image-2.5-Pro is geared towards premium image generation tasks, including detailed editing and precise text rendering within images. MAI-Voice-2-Flash, on the other hand, focuses on speed and cost-efficiency for applications like call centers and real-time speech services where large volumes are processed.

The deployment of these models has reportedly led to substantial savings across Microsoft's ecosystem. For instance, in PowerPoint, the use of MAI-Image-2.5 is said to reduce GPU costs by up to 84% compared to OpenAI's GPT-Image-2. In Dynamics 365 Contact Center, MAI-Voice-2-Flash has reportedly lowered GPU costs by as much as 89%.

This strategy of developing tailored models for specific use cases aims to achieve optimal performance and cost-effectiveness, catering to diverse needs from creative studios to large-scale customer service operations.