Maybe this is why OpenAI hides o1/o3 reasoning tokens - constraining output at inference time seems to be easy to implement for other models and others would im

Maybe this is why OpenAI hides o1/o3 reasoning tokens - constraining output at inference time seems to be easy to implement for other models and others would immediately start their photocopiers. It also gave them a few months to recoup costs!