Microsoft Challenges OpenAI With Faster, In-House 'MAI' Models

Microsoft Corp (NASDAQ:MSFT) is accelerating its shift toward in-house AI development with a new lineup of models designed to improve speed, efficiency, and cost across core enterprise use cases.

Speech And Transcription Capabilities Advance

Microsoft rolled out MAI-Transcribe-1 to deliver speech-to-text across the 25 most-used languages, processing batch workloads at 2.5 times the speed of its Azure Fast offering while maintaining strong real-world accuracy. The company said the model outperformed competing systems in benchmark tests across 11 languages and plans to integrate it into products such as Teams.

Voice And Image Models Target Scale And Creativity

MAI-Voice-1 enables developers to generate natural, expressive speech and create custom voices from short audio samples, producing 60 seconds of audio in 1 second while efficiently using the GPU. Alongside it, MAI-Image-2 boosts image generation speeds by at least two times while maintaining comparable quality. Microsoft designed the model for creative professionals, and companies like WPP are already using it at scale.

AI Strategy Focuses On Performance And Self-Sufficiency

Microsoft is deploying these models through Foundry and MAI Playground with competitive pricing, while embedding governance and safety controls for enterprise use. Microsoft AI CEO Mustafa Suleyman told Bloomberg on Thursday that the company is targeting state-of-the-art ...

Full story available on Benzinga.com