Yeah the efficiency gain is a big win! However for basic non-reasoning tasks it must surely make sense to use models like 4.1 still?
I might add I’ve been prototyping in Codex CLI with 5.5 and it absolutely flies through requirements 
2 Likes