The Redmond, Washington-based company has tested how the models, called MAI, might perform on a variety of tasks, including powering elements of the company’s Copilot-branded AI assistants.
Copilots are designed to handle a broad range of user questions, as well as provide specific suggestions for people working on documents or taking conference calls.
Microsoft is also developing so-called reasoning models, which are designed to field more complex queries and display human-like problem-solving, the person said. OpenAI and Anthropic, as well as Alphabet, are also plugging away on such models. Microsoft last month incorporated OpenAI’s o1 reasoning model into its Copilot products.
“As we’ve said previously, we are using a mix of models, which includes continuing our deep partnership with OpenAI, along with models from Microsoft AI and open source models,” a Microsoft spokersperon said.
See also: Musk's xAI seeks up to US$200 billion valuation in new funding
The Information reported on Microsoft’s models work earlier Friday.
Microsoft’s MAI models could eventually lessen the company’s reliance on OpenAI. Microsoft has backed the ChatGPT maker with some US$13 billion ($17.30 billion), and the relationship between the two companies has been a subject of debate in tech industry circles.
Microsoft and OpenAI recently renegotiated their deal, announcing in January that OpenAI could power its services with servers from other cloud-computing rivals, so long as Microsoft didn’t want the business itself. Their agreement runs until 2030.
See also: NCS commits $130 mil over three years to accelerate AI adoption across Asia Pacific
OpenAI declined to comment.
“We’re both successful when each of us are successful,” Microsoft CFO Amy Hood said at a Morgan Stanley conference earlier this week. “So as you go through that process, I do think everybody’s planning for what happens for a decade, or two decades. And that’s important for both of us to do.”
The company already offers a set of smaller, Microsoft-built models, called Phi, and resells AI models made by a wide range of other companies. It has tested how models from companies including Anthropic, DeepSeek, Meta Platforms and Elon Musk’s xAI would fare backing up Copilot, the Information reported.
“We feel great about having leading models from OpenAI, we’re still incredibly proud of that,” Hood previously said. “But we also have other models, including ones we build, to make sure that there’s choice.”
