What's your reasoning effort set to? Max now uses way more tokens and isn't suggested for most usecases. Even the new default (xhigh) uses more than the old default (medium).