Microsoft Just Released Three AI Models. That Is the Story, Not the Partnership.
For five years, the Microsoft-OpenAI relationship was described as a partnership. The contract told a different story. Microsoft was a distributor, a cloud host, a checkbook. It had invested more than $13 billion. It had the right to resell OpenAI models exclusively through Azure. It did not have the right to build its own frontier models, or to pursue AGI independently, or to partner with third parties on artificial general intelligence. Those clauses existed in writing. They were deleted in September 2025.
What replaced them is now visible.
On Thursday, Microsoft announced three in-house models through its Azure Foundry platform: MAI-Transcribe-1, a speech recognition model; MAI-Voice-1, a speech generation model; and MAI-Image-2, a text-to-image model. Pricing starts at $0.36 per hour for transcription, $22 per million characters for voice, and $5 per million tokens for image input. The image model debuted at number three on the Arena.ai leaderboard for its category. The transcription model ranks first on the FLEURS benchmark in 11 core languages and beats Whisper-large-v3 on the remaining 14. The voice model generates 60 seconds of audio in under one second on a single GPU. Mustafa Suleyman, speaking to VentureBeat, put it plainly: "We're now a top three lab, just under OpenAI and Gemini."
The three models were built by teams of roughly 10 people each. That is not a research laboratory. That is a product squad with a compute budget and a point to prove.
The backstory is the deal Microsoft made with itself. In October 2025, Microsoft acquired a 27 percent stake in OpenAI Group PBC, valued at approximately $135 billion at the time of signing. OpenAI committed to purchasing $250 billion in Azure services. The exclusivity clause was removed. Microsoft was no longer bound by contract from pursuing its own AGI. It had, in Suleyman's words, "the ability to independently pursue our own superintelligence." That is not a metaphor. That is a contractual change.
The $13 billion question is what took so long. Microsoft acquired Inflection AI for $650 million in March 2024 and hired its founder Mustafa Suleyman to lead Microsoft AI. The MAI Superintelligence team was formed and announced in November 2025. The three models shipped Thursday. The pipeline was always there. The contract was the only thing that kept it dormant.
The Foundry angle matters commercially. Azure Foundry is the same platform where Microsoft previously resold OpenAI models exclusively. Now it hosts Microsoft's own models alongside them. This is not a product refresh. It is a platform conflict that Microsoft has chosen to resolve in its own favor.
OpenAI, for its part, signed a $300 billion deal with Oracle in June 2024 and raised $110 billion at a $730 billion valuation in February from Amazon, Nvidia, and Softbank. It does not need Microsoft the way it once did. Microsoft, having rebuilt its own AI program around Inflection and Suleyman, no longer needs OpenAI the way it once did. The partnership survives on commercial terms. The strategic trust is gone.
What remains is a commodity. Two companies selling access to frontier models through a shared cloud platform, competing on price and performance rather than exclusivity. The models Microsoft released this week are competitive on benchmarks. The economics of the deal are not.
