Three models. One signal. On April 2, 2026, Microsoft AI released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, all built in-house, all available via Microsoft Foundry in preview. Silicon Republic confirmed the releases and noted their immediate integration into Microsoft’s existing product stack.
MAI-Transcribe-1 handles speech-to-text across 25 languages. Microsoft says it transcribes more than twice as fast as the company’s existing Azure Fast offering, according to Morningstar’s Dow Jones reporting. That speed comparison is Microsoft’s own claim, independent benchmarks haven’t been published. Specific accuracy figures aren’t confirmed from the sources available for this brief, so the accuracy story rests on Microsoft’s characterization of enterprise-grade performance across those 25 languages.
MAI-Voice-1 generates natural-sounding speech and lets Foundry users create custom voices from a few seconds of audio. That custom-voice capability is notable for enterprise applications, think branded voice interfaces, accessibility tools, and localized customer service. Microsoft says the model is fast enough for real-time applications, though specific generation speed figures from the primary source weren’t independently verifiable for this brief.
MAI-Image-2 is already deployed in production with enterprise partners including WPP, confirmed by Morningstar. That existing deployment matters: it’s evidence of enterprise validation before the public preview launch, not just a press release claim. The model targets enterprise use cases rather than consumer creative applications.
All three models power Microsoft products already in market. Copilot, Bing, and Azure Speech are running on these models now. That’s the distinction between this announcement and a typical AI product launch, the integration is live, not pending.
The strategic context is straightforward. Microsoft built these models internally rather than using OpenAI’s offerings for these capability areas. Silicon Republic described it as a move that “places the company in direct competition with enterprise AI rivals, despite its deep ties with OpenAI.” That framing is reasonable given the evidence: MAI-Transcribe-1 competes directly with transcription services that have existed independently of Microsoft’s OpenAI relationship, and MAI-Voice-1 enters a market that includes ElevenLabs, Cartesia, and others.
What’s missing from the public record matters here. No official Microsoft press release or blog post was included in the verified source set for this brief, the confirmed details come through journalism. That means the capability story is real but the specifics remain Microsoft’s claims as reported, not primary-source verified. Treat the performance figures as directionally accurate until independent evaluation publishes.
Watch for third-party evaluations on MAI-Transcribe-1 specifically. Enterprise transcription is a measurable capability with established benchmarks. If the accuracy and speed claims hold up independently, this becomes a strong Azure competitive story. If they don’t, the narrative around Microsoft’s AI independence strategy gets more complicated. The Foundry preview period is the window to watch.
Developers with existing Azure Speech integrations should evaluate MAI-Transcribe-1 for migration potential. The 25-language coverage and claimed speed advantage are worth testing against current workflows, keeping in mind that the benchmarks are Microsoft’s own at this stage.