Microsoft launched its first two homegrown AI models this week. The models are now available in various Copilot programs and could signal a move to start incorporating the company’s own models into the Windows-centric assistant instead of relying so heavily on OpenAI’s GPT models.
The new models, called MAI-Voice-1 and MAI-1-preview, debuted on Thursday, August 28. Microsoft shared details about the models in a blog post, where the company highlighted their capabilities. MAI-Voice-1 is now available in Copilot and Copilot Labs, and Microsoft says that it can be used to generate up to 60 seconds of audio in under one second, all while relying on a single GPU. Microsoft claims that this “lightning-fast” efficiency makes it “one of the most efficient speech systems available today.”
MAI-1-preview, on the other hand, is designed to provide consumer-level benefits by following instructions and “providing helpful responses to everyday queries.” It’s currently only being made available in LMArena, though Microsoft says it plans to eventually roll it out for specific text-based cases in Copilot as the weeks progress. For now, though, the company wants to look into improving it with user feedback.
A long way from replacing OpenAI
While Microsoft’s launch of its own in-house AI has major implications on the future of Copilot, we don’t expect Microsoft to ditch its partnership with OpenAI any time soon. There’s little doubt that having access to GPT’s latest models has helped Microsoft expand what Copilot can do exponentially without having to rely on building its own models from the ground up. It’s been a good tactic, as it has given the company tons of usage data with which to build its own models. Meanwhile, Microsoft has even added features to Copilot that ChatGPT doesn’t have yet.
This approach has given Microsoft’s AI division time to focus on specific models and improvements, and Microsoft even says that its in-house options aren’t designed for enterprise-level queries, something that it usually focuses heavily on for advertising Copilot’s productivity features. So, it would make sense for Microsoft to continue relying on GPT for that side of things, at least for now.
We’ll be curious to track these models in the months to come, as Microsoft has plenty of data to feed them already, and future input will only help the new models become even more capable. Since these models are focused on consumers, though, it’s possible we could eventually see these models make the leap to Copilot in Edge, too, as Microsoft tries to morph its Internet Explorer replacement into a proper AI browser.