According to a person engaged in the program, The Information revealed on Friday that Microsoft is creating its own artificial intelligence reasoning models, internally known as “MAI,” in a strategic move that indicates a departure from its sole dependence on to compete with OpenAI and may offer them to developers. And by creating its own potent AI models and looking for substitutes to power products like Microsoft’s Copilot bot, the company is stepping up its efforts to compete with OpenAI, a long-time partner.
The plan, which was initially revealed by The Information, aims to lessen Microsoft’s reliance on the ChatGPT manufacturer while improving the company’s AI skills.
According to the article, the Redmond, Washington-based business, which is a significant supporter of OpenAI, has started experimenting with models from xAI, Meta, and DeepSeek as possible OpenAI substitutes in Copilot.
According to The Information, Microsoft has created AI “reasoning” models of its own that are similar to OpenAI’s o1 and o3-mini models. Tensions between the companies are reportedly heightened by OpenAI’s alleged refusal to provide Microsoft with technical information on how o1 operates.
Even if its early collaboration with the firm placed Microsoft at the forefront of its Big Tech colleagues in the lucrative AI race, the company has been seeking to lessen its reliance on the ChatGPT creator.
In order to diversify from the existing OpenAI underlying technology and cut costs, the business has been working on integrating internal and third-party AI models to power its flagship AI product, Microsoft 365 Copilot, according to an exclusive December Reuters article.
One of the main selling points of Microsoft’s 2023 365 Copilot was that it made use of OpenAI’s GPT-4 model.
The Information story claims that Mustafa Suleyman, the head of Microsoft’s AI business, has finished training a family of models known internally as MAI that outperform the top models from Anthropic and OpenAI on well recognized benchmarks.
Also according to Bloomberg, Microsoft has also created a family of models known as MAI that rival OpenAI’s own and may make them available via an API later this year. Microsoft is reportedly exploring alternative AI models from xAI, Meta, Anthropic, and DeepSeek as potential OpenAI technology substitutes in Copilot in tandem with those initiatives.
According to the study, the company is also developing reasoning models that might directly compete with OpenAI’s. These models employ chain-of-thought procedures, a reasoning process that produces responses with intermediate reasoning abilities while tackling complicated issues.
According to the story, Suleyman’s team is already testing replacing the MAI models in Copilot with OpenAI’s models, which are far larger than an earlier family of Microsoft models known as Phi.
According to the source, the corporation is thinking of making the MAI models available as an application programming interface later this year so that independent developers may include them into their own programs.
Requests for response from Reuters were not immediately answered by Microsoft or OpenAI.
In its flagship AI product, Microsoft 365 Copilot, Microsoft has already started evaluating AI models from xAI, Meta, and DeepSeek as possible substitutes for OpenAI’s technology. Despite Microsoft’s substantial $13.75 billion investment in OpenAI since 2019, this diversification highlights the company’s larger goals in the AI field.
Microsoft, which has so far spent almost $14 billion in OpenAI, has sought to diversify its investments by appointing Mustafa Suleyman, a co-founder of DeepMind and Inflection, to head the tech giant’s AI initiatives.
Interestingly, Microsoft is currently testing the integration of these models into Copilot, sometimes taking the place of OpenAI’s technology. Later this year, the business may potentially provide MAI as an application programming interface (API), enabling other developers to integrate Microsoft’s AI features into their own software.
Microsoft’s decision to create AI internally is consistent with its larger plan to diversify its AI infrastructure. When Microsoft 365 Copilot was released in 2023, it was aggressively promoted as utilizing OpenAI’s GPT-4 model. But over time, Microsoft has looked at other AI models for three main reasons: to improve processing speeds, cut expenses, and prevent exclusive dependence.