Model-based systems engineering (MBSE) enables to verify the system performance using system behavior models, which can identify design faults that do not meet the stakeholders' requirements as early as possible, thus reducing the R&D cost and error risks. Currently, different domain engineers make use of different modeling languages to create their own behavior models. Different behavior models are verified by different approaches. It is difficult to adopt a unified integrated platform to support the modeling and verification of heterogeneous behavior models during the conceptual design phase. This paper proposes a unified modeling and verification approach supporting system formalisms and verification. The KARMA language is used to support the unified formalisms across MBSE models and dynamic simulations for different domain specific models. In order to describe the behavior model more precisely and to facilitate verification, the syntax of hybrid automata is integrated into KARMA. We implemented behavior models and their verification in MetaGraph, a multi-architecture modeling tool. Finally, the effectiveness of the proposed approach is validated by two cases: 1) the scenario of booking railway tickets using BPMN models; 2) the behavior performance simulation of unmanned vehicles using a SysML state machine diagram.