Comment by tifa2up

We tried GPT-5 for a RAG use case, and found that it performs worse than 4.1. We reverted and didn't look back.