As OpenAI partners with other corporations for large-scale investments, Microsoft has been developing a model to compete with OpenAI's AI model.
The technology-focused media outlet The Information reported on the 7th that Microsoft is developing its own artificial intelligence (AI) inference model to compete with OpenAI.
The AI team at Microsoft, led by Mustafa Suleyman, has recently completed training a group of AI models referred to internally as 'My' (MAI). Sources reveal that these models demonstrated performance nearly on par with the top AI models from OpenAI and Anthropic in benchmarks that assess AI capabilities.
In particular, this AI team is training inference models within the 'My' model group. This model utilizes the 'Chain-of-Thought' technique, which generates answers by going through intermediate reasoning steps when solving complex problems.
The Information conveyed that this inference model could compete directly with the OpenAI model.
OpenAI has been developing inference models of its 'o' series, which differ from general AI models. In September of last year, it launched the inference model 'o1' and released an upgraded small model 'o3' at the end of last January.
'My' is a much larger model than Microsoft's self-developed small language model 'Phi'. It is known that Microsoft is also testing the replacement of OpenAI's AI model with 'My' in its own AI productivity tool, 'Copilot'.
Such moves by Microsoft are interpreted as efforts to reduce dependence on OpenAI. Since 2019, Microsoft has invested $13 billion in OpenAI and has led the AI boom through a partnership for over two years.
However, Microsoft is now moving away from OpenAI technology to develop AI models in-house and incorporate models from other corporations.
Microsoft is also reportedly starting to test models from Elon Musk's AI startup xAI, Meta Platforms, the parent company of Facebook, and the Chinese AI startup DeepSeek.