Baidu launches DuMateBench benchmark for real-world AI agent delivery

Gartner has identified agent washing, where vendors falsely label scripted chatbots or RPA bots as AI agents. This marketing gap leads to significant risks, with Gartner predicting over 40% of agentic AI projects will be canceled by 2027. To address the need for genuine delivery, Baidu launched DuMateBench. This evaluation leaderboard tests AI agents across 200 office tasks in six categories, measuring tool use and continuous execution to ensure agents complete real-world tasks effectively.