TLDRs:
- Microsoft begins public testing of MAI-1-preview, aiming to reduce reliance on OpenAI.
- MAI-1 ranks 13th on LMArena, behind models from OpenAI, Google, and Anthropic.
- Microsoft recruited top AI talent from DeepMind and Inflection to accelerate MAI-1.
- Partnership-to-competition trend grows as Microsoft develops in-house AI capabilities.
Microsoft has officially launched public testing for its latest AI language model, MAI-1-preview, signaling a major step toward reducing its reliance on OpenAI for powering features like Copilot.
The model is currently available for evaluation on LMArena, a widely used AI benchmarking website.
According to Microsoft, the rollout of MAI-1-preview will be gradual, starting with select Copilot text features. Developers interested in early access can submit requests via an online form. The company emphasized that user feedback from this testing phase will be crucial to refining the model before broader deployment.
MAI-1 Performance in Benchmarks
On LMArena, MAI-1-preview currently ranks 13th for text tasks, trailing models from AI leaders such as OpenAI, Google, Anthropic, and xAI.
Despite this, Microsoft highlighted the technical scale behind the model, noting it was trained using 15,000 Nvidia H100 GPUs along with a cluster of Nvidia GB200 chips.
This level of computational power underscores Microsoft’s commitment to building a competitive AI solution internally, even as OpenAI continues to expand its user base, ChatGPT now reaches 700 million weekly users and the company is valued at around $500 billion.
From Partner to Competitor
The launch of MAI-1-preview illustrates a broader shift in the AI industry where strategic partners evolving into rivals. Microsoft and OpenAI initially formed a close partnership in 2019, with Microsoft investing $1 billion and becoming OpenAI’s exclusive cloud provider via Azure.
Excited to share our first @MicrosoftAI in-house models: MAI-Voice-1 and MAI-1-preview. Details and how you can test below, with lots more to come⬇️ pic.twitter.com/LtL2YmzuTv
— Mustafa Suleyman (@mustafasuleyman) August 28, 2025
Over five years, Microsoft has invested more than $13 billion while simultaneously developing competing AI models, officially listing OpenAI as a competitor in its annual reports.
This shift reflects OpenAI’s rapid growth and diversification, as it now distributes infrastructure needs across multiple cloud providers, including CoreWeave, Google, and Oracle, rather than relying solely on Microsoft’s Azure platform. Analysts suggest this transition from collaboration to competition is increasingly common in the AI sector, where early partnerships often serve as a springboard to direct rivalry.
Talent Acquisition Drives AI Development
Microsoft’s strategy to build MAI-1 also highlights the importance of strategic hiring in AI development. The company recruited Mustafa Suleyman from AI startup Inflection, along with several colleagues, and added approximately two dozen researchers from Google’s DeepMind in recent months.
Suleyman’s experience like co-founding DeepMind and later running Inflection as an OpenAI competitor, provides Microsoft with seasoned AI leadership and valuable industry networks. This “acqui-hiring” approach enables Microsoft to accelerate its in-house AI development, effectively compressing years of research and experimentation into a shorter timeline.
With MAI-1-preview, Microsoft now has its first foundation model trained entirely in-house, marking a significant milestone even after investing billions into OpenAI. The move positions Microsoft not just as a partner but as a direct competitor in the rapidly growing AI market.
