might as well rethink about their pricing on gpt-realtime. for a quick test, spoke to that model for ~5 minutes, ended up spending near $1.