OpenAI is trying to move as many requests as they can to a "smaller" model (still suspected to be ~200B).

I suspect it to be >1T, just without reasoning.