Comment by muzani

The data isn't there. Most would be out of date, and there's no incentive to train a LLM on this. OpenAI doesn't even do a lot of things well; their RAG, STT, TTS, voice, embeddings is below average, but it's convenient for people who don't want to look elsewhere.

The best would be looking to AI influencers. The good ones will be familiar with the common drawbacks and look to ways to evaluate the new models. Some do a weekly summary and such.