Microsoft Explores In-House AI Models for Copilot, Reducing Reliance on ChatGPT

Microsoft is shifting focus towards developing its own AI models for Copilot, aiming to reduce dependency on OpenAI’s ChatGPT while exploring new AI frontiers.

Microsoft, which was one of the first to back OpenAI, has always been quick to show off how its products, especially Copilot, play nice with the latest ChatGPT models. But here’s the twist: they’re now shifting gears, betting big on their own AI tech to power Copilot. Enter ‘MAI’—a new lineup of AI models Microsoft’s cooking up, aiming to go toe-to-toe with the big guns from OpenAI and Anthropic. (Because why not, right?)

Led by AI guru Mustafa Suleyman, Microsoft’s not just looking to cut the cord with OpenAI; they’re building their very own AI stack for Copilot. This comes hot on the heels of their latest small language models, Phi-4-multimodal and Phi-4-mini, which are pretty much the Swiss Army knives of AI—multi-modal, versatile, and ready to take on ChatGPT and Gemini. And guess what? Developers can get their hands on these through Azure AI Foundry, HuggingFace, and even the NVIDIA API Catalog. Talk about spreading the love.

Now, here’s where it gets juicy. Microsoft’s own tests show Phi-4 leaving Google’s Gemini 2.0 in the dust on several fronts, especially when it comes to summarizing speeches and keeping up with GPT-4o. The plan? Roll out the ‘MAI’ models on Azure, and while they’re at it, maybe flirt with third-party AIs like DeepSeek, xAI, and Meta to give Copilot that extra oomph.

But wait, there’s more. Microsoft isn’t stopping at just language models. They’re diving headfirst into reasoning AI models, setting the stage for a showdown with OpenAI’s GPT-o1 and newcomers like DeepSeek. This push comes amid some not-so-friendly vibes between Microsoft and OpenAI over tech sharing and transparency. With Phi-4 already showing off its smarts in language, math, and visual science reasoning, Microsoft’s proving it’s serious about pushing AI’s problem-solving chops to the next level. Because in the world of AI, it’s adapt or get left behind.