Do you recommend setting the Codex context size to 1M?

I'd assume models are optimized (RL'd on) the default parameters...?