Microsoft Unveils Cost-effective AI Models, Reducing Expenses by up to 89%
WHO:
VentureBeat - Microsoft AI's Superintelligence team announced new models.
VentureBeat - Rob Reilly, global chief creative officer at WPP, commented on the Pro model.
VentureBeat - Satya Nadella, Microsoft CEO, framed the announcements in a post on X.
WHAT:
VentureBeat - Microsoft AI released MAI-Image-2.5-Pro and MAI-Voice-2-Flash into public preview.
VentureBeat - MAI-Voice-2-Flash is designed for high-volume enterprise workloads and runs twice as fast as MAI-Voice-2.
VentureBeat - MAI-Image-2.5 is now the default for key image-editing scenarios in OneDrive, resulting in a 26% increase in save rates and 25% lower P95 latency.
VentureBeat - Microsoft claims a 50% relative reduction in transcription and language-identification error rates with MAI-Transcribe-1.5 in healthcare applications.
VentureBeat - Microsoft released two new in-house AI models: MAI-Image-2.5-Pro and MAI-Voice-2-Flash.
VentureBeat - MAI-Image-2.5-Pro is priced at $5 per million text input tokens, $8 per million image input tokens, and $106 per million image output tokens.
VentureBeat - MAI-Voice-2-Flash is priced at $15 per million characters and runs twice as fast as MAI-Voice-2.
VentureBeat - Microsoft claims GPU cost reductions of up to 89% with the new models.
Techmeme - Mustafa Suleyman stated that MAI models are approximately 85% cheaper to run in PowerPoint.
WHERE:
VentureBeat - The new models are integrated into Microsoft products including Bing, PowerPoint, OneDrive, Dynamics 365, Excel, GitHub Copilot, and Azure.
VentureBeat - MAI-Voice-2-Flash powers Dynamics 365 Contact Center, used by T-Mobile and EasyJet.
VentureBeat - Microsoft’s Dragon Copilot is used by 170,000 medical providers.
WHEN:
VentureBeat - The announcement was made on a Wednesday.
VentureBeat - MAI-Image-2.5-Pro recently launched at No. 2 for image editing on Arena.
VentureBeat - The Dragon Copilot processed 28 million patient encounters last quarter.
WHY:
VentureBeat - Microsoft aims to power its products without relying on OpenAI's models.
VentureBeat - The company is focusing on building purpose-built models internally to serve millions of users.
HOW:
VentureBeat - Microsoft is packaging its "hill-climbing" approach as a template for other AI-native, SaaS, or enterprise companies through Foundry and Frontier Tuning.
VentureBeat - Microsoft’s production metrics show in-house models cutting GPU costs by up to 89%.
VentureBeat - MAI-Image-2.5 is now the default for key image-editing scenarios in OneDrive.
VentureBeat - Microsoft employs a "hill-climbing machine" methodology to improve model performance and efficiency.
(UPDATED July 24, 2026 05:53 UTC)
Microsoft has released MAI-Image-2.5-Pro and MAI-Voice-2-Flash into public preview, with MAI-Voice-2-Flash designed for high-volume enterprise workloads and operating twice as fast as its predecessor. The new MAI-Image-2.5 has become the default for key image-editing scenarios in OneDrive, leading to a 26% increase in save rates and a 25% reduction in P95 latency. Additionally, Microsoft reports a 50% relative reduction in transcription and language-identification error rates with MAI-Transcribe-1.5 in healthcare applications, while also offering its "hill-climbing" approach as a template for other AI-native companies through Foundry and Frontier Tuning.
--
(UPDATED July 24, 2026 01:31 UTC)
Microsoft has unveiled two new in-house AI models, MAI-Image-2.5-Pro and MAI-Voice-2-Flash, which are designed to enhance its product offerings while significantly reducing costs. The MAI-Image-2.5-Pro is priced at $5 per million text input tokens and is now the default for key image-editing scenarios in OneDrive, while MAI-Voice-2-Flash runs twice as fast as its predecessor and costs $15 per million characters. These models are integrated into various Microsoft products, including Bing and Dynamics 365, and the company aims to power its services independently without relying on OpenAI's technology.