Report: Microsoft’s New MAI-1 Model May Challenge OpenAI’s Leadership

Microsoft is reportedly readying its largest large langage model (LLM) yet that could compete with the best AI models from OpenAI, Google, and Anthropic. According to a report from The Information (paywalled), the development of Microsoft’s new AI model, which is codenamed MAI-1, is being led by DeepMind and Inflection co-founder Mustafa Suleyman, who recently became CEO of Microsoft AI.

Citing two Microsoft employees familiar with the matter, the report says that MAI-1 will have approximately 500 billion parameters and “be far larger than any of the smaller, open source models that Microsoft has previously trained.” In comparison, OpenAI’s GPT-4 model is said to have over 1 trillion parameters, while the largest version of Meta’s Llama 3 AI model only has 70 billion parameters.

According to The Information, Microsoft’s new MAI-1 model is being developed entirely in-house. It’s not leveraging existing models from Inflection, Suleyman’s previous AI startup, though “it may build on training data and other tech from the startup.” Anyway, MAI-1 will be much more ambitious (and expensive) than the Phi-3 Mini small language model Microsoft announced two weeks ago.

As Microsoft CEO Satya Nadella explained in March, the purpose of Microsoft’s new AI division, which is led by high-profile outsiders in the AI field such as Suleyman, is to “build world-class AI products” that leverage OpenAI technology. “We will continue to build AI infrastructure inclusive of custom systems and silicon work in support of OpenAI’s foundation model roadmap, and also innovate and build products on top of their foundation models,” Nadella said at the time.

Microsoft invested over 10 billion dollars into OpenAI to get an exclusive license on GPT-4 and all of the company’s other AI models. However, Microsoft’s deal with OpenAI is already being scrutinized by the US FTC, the UK’s CMA, and the EU Commission, and this may be why the company appears to be hedging its bets.

In addition to developing its own AI models, Microsoft also announced a strategic partnership with French-based AI startup Mistral AI in February. However, this partnership is already being scrutinized by the EU Commission. Again, it may be in Microsoft’s best interest to develop its own AI models, and even though the main purpose of MAI-1 is still unclear, The Information believes that Microsoft could officially announce the new AI model at its Build conference later this month.