GPT-4 via the API is better, but what you really want is LLaMa2 Chat 70B. This will get further aligned and eventually dominate.

Meta might release a bigger model if 70B stalls out, they seem to be drip feeding the leaks to dethrone OpenAI.