Microsoft launches two new in-house AI models, cutting GPU costs up to 89%

Microsoft ↗ 2.43% on Wednesday unveiled MAI-Image-2.5-Pro and MAI-Voice-2-Flash, its latest in-house AI models, as the company accelerates a strategy to replace third-party systems from OpenAI and Anthropic across its product lineup. The announcement, made through the company's AI division, showed production deployments already achieving GPU cost reductions of up to 89% compared with OpenAI's models in some workloads.microsoft+1

The two new models join a family of seven MAI models first introduced at the Build developer conference in June, marking what the company calls a shift toward "long-term self-sufficiency" in AI.enterprisedna+1

Deployed Across Core Products

According to Microsoft's official blog post, MAI models are now running in production across Bing Image Creator, PowerPoint, OneDrive, Dynamics 365, and Azure. Bing Image Creator is now "100% in-house," powered entirely by MAI-Image-2.5. In PowerPoint, the image model reduced GPU costs by up to 84% compared with OpenAI's GPT-Image-2, while in Dynamics 365 Contact Center — used by customers including T-Mobile and EasyJet — MAI-Voice-2-Flash cut GPU costs by up to 89%.kucoin+1

Separately, Bloomberg reported on July 7 that Microsoft had begun routing tens of thousands of Copilot prompts per week in Excel and Outlook to MAI models instead of OpenAI and Anthropic. Microsoft AI CEO Mustafa Suleyman told Bloomberg: "We pay a lot of money to Anthropic, so our goal is to reduce and ultimately eliminate that cost".channelinsider+2

A Broader Self-Sufficiency Push

The MAI family spans reasoning, coding, image generation, voice, and transcription. MAI-Code-1-Flash is already integrated into GitHub Copilot and Visual Studio Code, while MAI-Thinking-1, the company's first reasoning model, is available through Microsoft Foundry. Microsoft says the models were trained from scratch on commercially licensed data without distillation from third-party systems.geekwire+2

Still, the shift is incremental. Bloomberg's reporting noted that MAI models handle only a small percentage of Microsoft's overall AI traffic, with complex tasks still routed to OpenAI's frontier models. Microsoft continues to maintain its partnership with OpenAI, in which it has invested roughly $13 billion.thecooldown+2

Cost Pressure Drives the Pivot

The move reflects mounting pressure across the industry to manage AI inference costs as usage scales. By running its own models on Azure infrastructure and its custom Maia silicon, Microsoft avoids per-token fees to outside providers. MAI-Voice-2-Flash is priced at $15 per million characters — 32% cheaper and twice as fast as its predecessor. VentureBeat reported that the production cost data may prove more consequential than the model launches themselves, signaling a template other enterprises may follow.magicshot+3

"Each of these enhancements is a step toward the same goal: Microsoft products, powered by Microsoft models, built to serve the people who use them," the company wrote.microsoft