The Microsoft AI Superintelligence team presented two models of its own development: the MAI–Image-2.5-Pro image generator and MAI–Voice-2-Flash — a voice model for large-scale corporate workloads.
The Microsoft AI Superintelligence team presented two models of its own development: the MAI–Image-2.5-Pro image generator and MAI–Voice-2-Flash — a voice model for large-scale corporate workloads.
MAI–Image-2.5-Pro targets the premium segment: promotional illustrations, detailed editing, and accurate display of text inside images, VentureBeat writes .
Microsoft has set a price of $5 per million input text tokens, $8 per million input image tokens, and $106 per million output image tokens.
MAI–Voice-2-Flash is twice as fast as MAI–Voice-2 and costs 32% less—$15 per million characters. It is designed for the routine but huge mass voice processing market: call centers, voice agents, and real-time speech applications, where latency and call cost outweigh minimal benefits in emotional expressiveness.
Microsoft's production figures show that its own designs reduce GPU costs by up to 89%.
The model releases themselves are second only to the indicative deployment metrics that Microsoft has added, which are seen as a compelling justification for pushing third-party models out of its entire product line.
Yes, Bing Image Creator now runs 100% on MAI–Image-2.5 — this is the first time the company's consumer graphics service has completely switched to its own "engine."
In PowerPoint, Microsoft says MAI–Image-2.5 reduces GPU overhead by up to 84% compared to GPT-Image-2, OpenAI’s image generation model. And in OneDrive, where MAI–Image-2.5 has become the standard for core editing scenarios, the company has seen a 26% increase in saves, a reduction in P95 latency of about 25%, and a 2.5x performance improvement on average workloads.
On the voice side, MAI–Voice-2-Flash now powers Dynamics 365 Contact Center, a platform used by customers such as T-Mobile and EasyJet. Microsoft claims that this system reduces GPU costs by up to 89%. The model has also been integrated into Azure Voice Live for developers building conversational voice agents.
