@thsottiaux A few hours ago, right after the reset, it was really smart and efficient. A few hours later, it started producing low-quality output and consuming a lot of tokens. Is this high consumption due to a drop in intelligence causing random tool usage and testing?