Probably not? At least not any public models I've heard of. If you had a model architecture design AI that was close to that good, you'd want to keep the secret sauce to yourself and use it to publish other research or develop products.

LLMs show absolutely huge potential for being a conductor or executive that coordinates smaller modules. The plug-ins coming to ChatGPT are the more traditional software version of this. How long until an LLM can determine it needs a specific kind of machine learning model to understand something and just cooks up and architecture and can choose appropriate data?