Microsoft just put two new AI models into public availability: MAI-Image-2.5-Pro, built for high-quality image generation and editing, and MAI-Voice-2-Flash, built for fast and natural-sounding voice in call centers and voice-driven software. Neither is experimental. Both are already running inside Microsoft products at scale.
The image model is now the default engine inside Bing's image creation tool, PowerPoint's image features, and OneDrive's photo editing. The voice model now powers the call center software used by T-Mobile, EasyJet, and others through Microsoft's Dynamics 365 platform. These are real deployments, not pilots.
What makes this more than a product update is the cost data Microsoft published alongside it. Running the image model in PowerPoint cut computing costs by up to 84% compared to the OpenAI model it replaced. The voice model reduced those costs in the call center software by up to 89%. In OneDrive, users saved their edits 26% more often after the switch, and the system became noticeably faster. Better product, lower cost: that is a difficult combination to argue against internally.
To understand why this matters, some context helps. Microsoft has invested around 13 billion dollars in OpenAI since 2019, and for most of that period its AI products ran almost entirely on OpenAI's technology. That arrangement gave Microsoft a head start in AI, but it also meant paying for every query, with no control over pricing, timing, or model direction. A renegotiated deal in late 2025 gave Microsoft the legal freedom to build competing models, and it has moved quickly since then.
At its developer conference in June 2026, Microsoft unveiled seven in-house models covering image generation, voice, transcription, reasoning, and code. The two models launched this week are additions to that family: a higher-quality image option and a faster, cheaper voice option. The strategy is to offer a range, so that each product can pick the right balance of quality, speed, and cost for its specific job.
For businesses that use Microsoft's tools, the immediate effect is mostly invisible. The AI inside PowerPoint or a customer service phone line just works better and faster. But there are two things worth watching.
First, if you use Microsoft's Dynamics 365 for contact center operations, your voice AI has already changed. The new model supports more than 15 languages and is designed specifically for high-volume, real-time phone conversations. The quality difference in voice naturalness is real, and for customer-facing operations it affects how your brand sounds.
Second, Microsoft's medical transcription model, used by 170,000 doctors and processing 28 million patient encounters per quarter, has also switched to an in-house model across 58 languages. It cut transcription errors in half in internal tests. If your business touches healthcare administration or medical records, this is the same technology stack you will likely encounter.
The broader shift is that Microsoft is no longer simply a company that sells access to other people's AI. It is becoming a company that builds and owns the AI inside its own products. OpenAI and Anthropic still handle significant traffic across Microsoft's platforms, and that will not change overnight. But the direction of travel is clear, and the cost savings make it self-reinforcing: every feature Microsoft moves in-house makes the economics of staying in-house more attractive.