what LLM model have good enough latency/good quality? I'm still on gpt 4o and have no idea about latency impact of moving to a smarter more recent model