Software in the cloud is characterised by the need to be highly adaptive and continuously available. Incremental changes are applied to the deployed system and need to be tested in the field. Different configurations need to be tested. Higher quality standards regarding both functional and non-functional properties are put on those systems, as they often face large and diverse customer bases and/or are used as services from different peer service implementations. The properties of interest include interoperability, privacy, security, reliability, performance, resource use, timing constraints, service dependencies, availability, and so on. This paper discusses the state of the art in model-based testing of cloud systems. It focuses on two central aspects of the problem domain: (a) dealing with the adaptive and dynamic character of cloud software when tested with model-based testing, by developing new online and offline test strategies, and (b) dealing with the variety of modeling concerns for functional and non-functional properties, by devising a unified framework for them where this is possible. Having discussed the state of the art we identify challenges and future directions.