All those apply to OpenAI+Codex too, but they're far more generous with limits than Anthropic, and with granting fresh limits to apologize when they fuck up.