Microsoft can use OpenAI models but it's not the model that's the problem, it's the application of them. Anthropic simply knows how to execute better.

Microsoft has a model nearly as old as the company.

Attempt to build a product... Fail.

Buy someone else's product/steal someone else's product... Succeed.

They do have some in-house LLM's (Phi) but they seem to either have issues with, or not thinking it's worth it, to develop large flagship ones.

One has existed since the 80s, when was the other founded?

A large language model, or a business model?