Researchers found reasoning inefficiencies in open-weight models, with token usage often exceeding closed counterparts by wide margins.| Digital Information World