Microsoft launches new in-house AI models it says cut costs up to 89% versus OpenAI

Microsoft AI released two new in-house models into public preview on Wednesday, MAI-Image-2.5-Pro and MAI-Voice-2-Flash, and published production data showing its models cut GPU costs by up to 89% compared with OpenAI alternatives, the company announced. The announcement was made by Microsoft AI's Superintelligence team roughly a year after the company committed to building purpose-built models internally. MAI-Image-2.5-Pro targets premium image generation with precise text rendering and is priced at $5 per million text input tokens, $8 per million image input tokens, and $106 per million image output tokens. MAI-Voice-2-Flash runs twice as fast as its predecessor at 32% less cost, priced at $15 per million characters, and is designed for high-volume voice applications such as call centers. The models now power Bing, PowerPoint, OneDrive, Dynamics 365, Excel, GitHub Copilot, and Azure. Microsoft said Bing Image Creator now runs entirely on MAI-Image-2.5, while PowerPoint sees up to 84% GPU cost reduction versus OpenAI's GPT-Image-2. Dynamics 365 Contact Center, used by T-Mobile and EasyJet, shows up to 89% GPU cost reductions with MAI-Voice-2-Flash. Dragon Copilot, used by 170,000 medical providers, now runs on MAI-Transcribe-1.5 with a reported 50% relative reduction in transcription and language-identification error rates across 58 languages. CEO Satya Nadella wrote in a post titled "Frontier Diffusion & Control" that Microsoft can now take saturated frontier capabilities and deliver them at scale and lower cost through models optimized for high-usage products while continuing to use frontier models for frontier needs. He said the company is beginning to route traffic across first-party surfaces to MAI models whenever they match or outperform frontier alternatives.