We're finding that nothing but the top ChatGPT model will do, but then we struggle with some latency.

gpt mini 4.1 or gpt 4.1 seem fine. also openai dropped new ones with similar pricing, i think it was 5.3

optimized for latency or deep thinking?

It's better to stick with the old models, these 5 has very high latency

I wish it was easier to use the Claude models. We're finding that they can't search the knowledge base.