Good point – the fine print states we're actually counting model evals

→ View original post on X — @jeremyphoward