Microsoft launched three in-house AI models, MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, available immediately through Microsoft Foundry, with the transcription model claiming the lowest word error rate across 25 languages on the FLEURS benchmark, beating OpenAI’s Whisper and Google’s Gemini Flash on most languages tested. The models were each built by teams of fewer than 10 engineers and run on half the GPU footprint of comparable competitors, with Microsoft pricing them below every major cloud rival to reduce its own infrastructure costs across Teams, Copilot, Bing, and PowerPoint. The launch follows a renegotiated contract with OpenAI that, until October 2025, had prohibited Microsoft from independently pursuing AGI, with Microsoft AI CEO Mustafa Suleyman telling VentureBeat the company’s goal is to be “completely independent” across all model modalities.
