“OpenAI reportedly used a 10x compute multiplier for post-training on its 'O3' model compared to its 'O1' model.”