Microsoft’s new AI model may pose a challenge to Google Gemini and GPT-4

  • Microsoft is developing MAI-1, a new AI language model that could rival leading models from Google, Anthropic, and OpenAI.
  • This venture marks Microsoft’s first major in-house AI development since its $10 billion investment in OpenAI.
  • Led by Mustafa Suleyman, the project leverages both new and previous AI techniques, following Microsoft’s acquisition of AI startup Inflection.
  • MAI-1 is projected to have about 500 billion parameters, exceeding Microsoft’s earlier models and potentially competing with OpenAI’s GPT-4.
  • The model represents Microsoft’s strategy to develop both small-scale mobile and large-scale cloud-based AI models.
  • The full application of MAI-1 is yet to be determined, with its deployment dependent on its performance in tests.

Main AI News:

According to The Information, Microsoft is developing a new AI language model named MAI-1, which could rival top-tier models from Google, Anthropic, and OpenAI. This is Microsoft’s first in-house AI model of this scale since investing more than $10 billion in OpenAI to reuse the startup’s models. GPT-4, OpenAI’s model, powers Microsoft Copilot and ChatGPT.

Mustafa Suleyman, previously a Google AI leader and former CEO of Inflection (an AI startup), is leading the MAI-1 project. Microsoft acquired Inflection’s staff and intellectual property for $650 million in March. Despite leveraging techniques from Inflection, MAI-1 is reportedly a completely new large language model (LLM), according to two Microsoft employees knowledgeable about the project.

MAI-1, which reportedly has around 500 billion parameters, is significantly larger than previous Microsoft open-source models (such as Phi-3, which we recently reported on) and requires increased computing power and data. This model is expected to rival GPT-4 (which may contain over 1 trillion parameters) and surpass smaller models from Meta and Mistral, which have 70 billion parameters.

MAI-1 indicates Microsoft’s dual AI strategy, targeting both small, locally operated models for mobile devices and cutting-edge cloud-powered models. Apple is reportedly exploring a similar direction. This reflects Microsoft’s willingness to explore AI independently from OpenAI, whose technology underpins its most ambitious AI features, including Windows’ chatbot.

The exact purpose of MAI-1 remains undetermined, as its best use will depend on performance, according to an insider quoted by The Information. Microsoft is using Nvidia GPU clusters and data from OpenAI’s GPT-4 alongside publicly available internet data for training purposes.

Progress over the next few weeks could lead to a preview of MAI-1 at Microsoft’s upcoming Build developer conference later this month, as noted by one of The Information’s sources.


Microsoft’s development of the MAI-1 model signifies a significant shift towards internal AI innovation, possibly reducing reliance on external technologies like those provided by OpenAI. This move could intensify competition among tech giants such as Google, Apple, and Meta, particularly in the realms of cloud and mobile AI solutions. The substantial investment in this new model suggests Microsoft’s commitment to leading in cutting-edge AI technology, potentially influencing market dynamics by setting new standards for AI capabilities and applications. If successful, MAI-1 could reshape competitive strategies across the technology sector, fostering advancements in AI that could ripple across various industries.