I thought o1 was a fine-tune of GPT-4o. I don't think o3 is though. Likely using the same techniques on what would have been the "GPT-5" base model.