We are rolling back an o4-mini snapshot that we deployed less than a week ago and intended to improve the length of model responses, because our automated monitoring tools detected an increase in content flags.