Microsoft Launches Three New Foundation Models Including MAI-Transcribe-1

Apr 3, 2026

Microsoft AI announced three new foundation models on April 2, 2026, developed by its MAI Superintelligence team formed in November 2025 under CEO Mustafa Suleyman. MAI-Transcribe-1 offers speech-to-text transcription in 25 languages at 2.5 times the speed of Azure Fast for $0.36 per hour, MAI-Voice-1 generates 60 seconds of audio per second starting at $22 per million characters, and MAI-Image-2 costs $5-$33 per million tokens. Microsoft positions these models as cheaper than Google and OpenAI alternatives, despite having invested over $13 billion in OpenAI and hosting OpenAI models through a multi-year partnership recently renegotiated to allow Microsoft's independent superintelligence research.

Microsoft's Three New Foundation AI Models and Capabilities

▪Microsoft AI announced the release of three foundational AI models on Thursday, April 2, 2026.
▪MAI-Voice-1 allows users to create a custom voice.
▪MAI-Image-2 was originally released on MAI Playground on March 19, 2026.
▪MAI-Transcribe-1 transcribes speech across 25 different languages into text.
▪MAI-Transcribe-1 and MAI-Voice-1 are available in MAI Playground.
▪Microsoft's MAI Superintelligence team was formed and announced in November 2025.
▪All three Microsoft AI models are being released on Microsoft Foundry.
▪MAI-Voice-1 allows users to generate 60 seconds of audio in one second.
▪MAI-Transcribe-1 is 2.5 times faster than Microsoft's Azure Fast offering.

Pricing Strategy and Market Competition

▪MAI-Voice-1 starts at $22 per 1 million characters.
▪MAI-Image-2 starts at $33 for 1 million tokens for image output.
▪MAI-Image-2 starts at $5 for 1 million tokens for text input.
▪MAI-Transcribe-1 starts at $0.36 per hour.

Microsoft's AI Strategy and OpenAI Partnership

▪Microsoft has invested more than $13 billion into OpenAI.

Perspective of Microsoft AI

▪Microsoft AI emphasizes that MAI-Transcribe-1 delivers 2.5 times faster performance compared to Azure Fast.
▪Microsoft AI positions MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 as cheaper alternatives to models from Google and OpenAI.