Comment by SOLAR_FIELDS

Imagine a tricky distributed systems problem where you feed all of the context of your setup to the LLM and it uses the advanced reasoning to diagnose possible avenues. I did that recently with a frontier model to unwrap some very tricky istio related connection pooling issues causing syn/ack floods.

For coding I usually use the fast frontier model like o4minihigh, but I bust out the fancy research models when I want things like general architecture and design feedbacks that require broader advanced reasoning