The move is part of a broader strategy to gain strategic independence from OpenAI, led by Mustafa Suleyman, aiming to reduce dependence while preserving existing partnerships.

Microsoft’s strategy centers on diversifying AI capabilities and reducing reliance on OpenAI by deploying proprietary tech across products and services.

The announcements follow a March organizational reshuffle under Satya Nadella, with Suleyman shifting to frontier model development and superintelligence instead of day-to-day Copilot oversight.

Microsoft unveils three in-house AI models—MAI-Transcribe-1 for multilingual, robust speech-to-text; MAI-Voice-1 for fast, realistic text-to-speech with customizable voices; and MAI-Image-2 for automated image generation aimed at both creative and enterprise use.

MAI-Transcribe-1 claims the lowest average word error rate on the FLEURS benchmark among its supported languages and outperforms OpenAI’s Whisper-large-v3 and Google Gemini 3.1 Flash on specified tests, with transcription pricing starting at $0.36 per hour.

MAI-Image-2 is broadly available, priced at $5 per million tokens for text input and $33 per million tokens for image output, and is being rolled out across Copilot, Bing Image Creator, and PowerPoint, with WPP among early enterprise adopters.

Pricing and in-house development could lower per-query costs for Copilot and related products, serving as a strategic hedge as investors seek returns on AI infrastructure.

Microsoft maintains its multi-year partnership with OpenAI and has invested heavily in AI research, including producing its own chips while also purchasing from external providers.

The push signals a narrative of competing with OpenAI, Google, and others, with Suleyman framing the launch as the first delivery from a “superintelligence team” formed six months earlier.

Microsoft emphasizes a lean development approach, with each model built by teams of fewer than 10 engineers, citing efficiency and cost advantages over large-scale R&D.

The company frames the rollout as a competitive response to market dynamics, suggesting pricing pressure and faster feature development across AI modalities for end users.

There is a roadmap toward a fully independent frontier model, broader developer access, ongoing evaluation of third-party options, and potential regulatory considerations through 2030.