I dont know if this is causing the difference in your experience, but Microsoft puts a lot of guardrails around their GPT deployments to avoid giving "harmful"

I dont know if this is causing the difference in your experience, but Microsoft puts a lot of guardrails around their GPT deployments to avoid giving "harmful" answers, so you do often get worse results than you do asking OpenAI's deployments the same question.