TokenX

By default, every task in a Codex session runs on whichever model you picked - the same deep reasoning for a one-line typo fix as for a concurrency bug. TokenX chooses per task instead.