OpenAI's model is closed source. IDK if distilling can be done via the API effectively? DeepSeek already has distilled models from other open source models like Qwen which have been done by 3rd party researchers, and I assumed that happened rapidly because they are all open source.