"OpenAI continues to be our partner on frontier models": Microsoft is open to using a range of AI models in 365 Copilot, but OpenAI remains its go-to choice

Microsoft has clarified its stance on the models that will power its 365 Copilot service after reports suggested it was considering integrating third-party alternatives to OpenAI’s model range.

Speculation around the two companies’ relationship emerged after a Reuters report last month claimed Microsoft had been “working on adding internal and third-party artificial intelligence models to power its flagship AI product Microsoft 365 Copilot.”

Sources said the decision was driven by Microsoft’s desire to diversify its current roster of models from OpenAI to minimize its reliance on the firm, largely due to concerns around cost and performance for enterprise customers.

Speaking to Reuters, a Microsoft spokesperson suggested OpenAI would remain Microsoft’s primary partner for frontier models, the most advanced models that underpin the rest of the generative AI ecosystem.

They added that Microsoft would "incorporate various models from OpenAI and Microsoft depending on the product and experience.”

ITPro approached Microsoft for clarification on how OpenAI would feature in future Copilot 365 products.

A Microsoft spokesperson reiterated the firm’s continued partnership with OpenAI on the underlying foundation models, but noted its existing agreement allows Microsoft to “go beyond” simply customizing OpenAI’s models.

“OpenAI continues to be our partner on frontier models. Our agreement allows us to go beyond fine-tuning and customize OpenAI’s models and IP precisely for our own needs so we can integrate them directly into our products,” they explained.

This leaves open the possibility for models from other sources to be used in Microsoft's products, and the spokesperson noted the firm has been clear that it will look at a range of sources for its generative AI solutions.

“As we have said before, we incorporate various models from OpenAI and Microsoft depending on the product and experience.”

In an example, Microsoft said some models perform better in terms of speed and accuracy than others in specific areas, and as such it would be using a range of models in its products to achieve the best results.

“For example, recommending specific edits to a paragraph in Word may be best suited for one model whereas formatting updates could be powered just as well—and perhaps more quickly—with a different model.”

Microsoft OpenAI partnership reaches $100 billion AGI definition

Microsoft and OpenAI have maintained close ties for several years now, with the latter’s AI models now woven throughout a raft of key Microsoft product ranges.

In 2020, OpenAI licensed its GPT-3 model exclusively to Microsoft, which gave the giant the right to integrate OpenAI’s models into its products and services, including Microsoft Office, Teams, and its Azure AI tools.

In turn, Microsoft became OpenAI’s exclusive cloud provider, allowing it to leverage Microsoft’s powerful Azure cloud infrastructure to deliver OpenAI’s products and API services.

The partnership also included collaboration on research into AI governance, ethics, and artificial general intelligence (AGI).

Microsoft launched its Azure OpenAI Service, designed to give enterprises access to OpenAI’s models as well as a raft of security and governance features in 2021.

Since its initial $1 billion investment in the generative AI pioneer, Microsoft’s total investment in OpenAI totals approximately $13 billion.