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

Microsoft AI released two in-house models into public preview on Wednesday: MAI-Image-2.5-Pro, its highest-fidelity image generator to date, and MAI-Voice-2-Flash, a speech model aimed at high-volume enterprise workloads. The company says the new models can cut costs by up to 89% versus OpenAI.

MAI-Image-2.5-Pro is positioned as Microsoft AI’s top image model for fidelity. MAI-Voice-2-Flash is built for speech at enterprise scale, where volume and unit cost matter more than one-off demos. Microsoft is also publishing production data alongside the preview, tying the launch to measured cost claims rather than only model cards.

For engineers and builders, an in-house image stack and a voice model designed for high-volume loads matter when image generation or speech is already in the product path and OpenAI-backed spend is a line item. Public preview means teams can test integration, latency, and quality against existing OpenAI-based flows before any production cutover.

The launch puts Microsoft’s own MAI models next to OpenAI in the stack Microsoft still partners with. Cost claims of up to 89% versus OpenAI set a direct price comparison for image and speech workloads, not a vague platform story. That frames the choice as own-stack versus partner models on unit economics and product fit.

Watch how MAI-Image-2.5-Pro and MAI-Voice-2-Flash perform in real preview use, and whether the published production data supports the up-to-89% cost claim on the workloads you actually run. Next step for most teams is a side-by-side trial on fidelity, speech quality, volume handling, and bill impact before changing default providers.