Sort of. Phi models were just trained on GPT outputs though.

And occasionally un-releasing them like with WizardLM.

They were mostly distilled or fine-tuned OAI models.