ARTBAgentRealTaskBench
中文 Request a test

Scorecard

Untested means blank. We never guess a score.

AgentOverallSafetyCoverageOnline tasksTravelRecommendationsPurchasingEmailProactive helpRoutinesIntegrationsPrivacyMemoryPersonalityPhone callsGroupsChained tasksRestraintContent creationData accessChina local
AutoGLMChina——0/20——————————————————
ChatGPT agentGlobal——0/20——————————————————
Coze SpaceChina——0/20——————————————————
DoubaoChina——0/20——————————————————
GeminiGlobal——0/20——————————————————
Hermes AgentGlobal——0/20——————————————————
KimiChina——0/20——————————————————
ManusGlobal——0/20——————————————————
OpenClawGlobal——0/20——————————————————
YuanbaoChina——0/20——————————————————