[fvv]

imo , the main feature is /fast ... who use 1M context and for what? the model become dumber already at 200K.. it's better to manage the context , and since 5.3, codex is very good at managing it