Mustafa Suleyman, CEO of Microsoft AI, speaks at an occasion commemorating the fiftieth anniversary of the corporate at Microsoft headquarters in Redmond, Washington, on April 4, 2025.
David Ryder | Bloomberg | Getty Images
Microsoft has largely relied on OpenAI’s synthetic intelligence fashions to energy AI options in its key merchandise. It’s attempting to reduce that dependence.
The software program firm stated Thursday that it is begun publicly testing a homegrown AI model that might lead to enhancements to its Copilot assistant for customers. The MAI-1-preview model is being examined on LMArena, an internet site the place folks can conduct evaluations.
“We will be rolling MAI-1-preview out for certain text use cases within Copilot over the coming weeks to learn and improve from user feedback,” Microsoft stated in a blog post. The firm has printed a kind the place builders can request early entry.
Meanwhile, Microsoft stays a key backer of OpenAI and strategic companion to the AI startup that is now valued at about $500 billion. Microsoft has invested over $13 billion in OpenAI, which in flip depends on cloud infrastructure from Microsoft to run its fashions.
Microsoft attracts on fashions from OpenAI to energy options in Bing, the Windows 11 working system and different merchandise.
But the businesses are clearly transferring into competitors in numerous methods. Last 12 months, Microsoft added OpenAI to the record of rivals in its annual report. It’s a roster that for years has included megacap friends Amazon, Apple, Google and Meta.
And in current months, OpenAI has turned to other cloud providers, equivalent to CoreWeave, Google and Oracle, to meet heavy demand. The startup’s ChatGPT assistant now reaches 700 million people per week.
On LMArena, Microsoft’s new model was ranked thirteenth for textual content workloads on Thursday, beneath fashions from Anthropic, DeepSeek, Google, Mistral, OpenAI and xAI.
Microsoft stated in its weblog publish that the model was refined with assist from round 15,000 of Nvidia’s H100 graphics processing items, and it additionally has a working cluster of Nvidia GB200 chips.
“We have big ambitions for where we go next — model advancements, an exciting roadmap of compute, and the chance to reach billions of people through Microsoft’s products,” stated Mustafa Suleyman, CEO of Microsoft’s AI unit, in a post on X.
Microsoft has beforehand developed small open-source language fashions below the title Phi.
MAI-1-preview represents “our first foundation model trained end to end in house,” Suleyman wrote on X.
Suleyman used to compete with OpenAI at startup Inflection. Last 12 months, Microsoft hired him and plenty of of his Inflection colleagues. Before that, Suleyman was a co-founder of DeepMind, an AI analysis startup that Google purchased in 2014.
Suleyman’s group inside Microsoft has been increasing, with about two dozen people coming from Google’s DeepMind AI lab in current months.
WATCH: OpenAI CFO Sarah Friar: Biggest issue we face is being ‘constantly under compute’
