Theory of Mind (ToM) is the ability to infer the mental state of others and is thus of pivotal importance to collaboration. Consequently, AI research has attempted to model the same capabilities in artificial agents. Recently, early attempts at modelling ToM have been called into question for modelling learning ToM in a task-based manner, even though ToM in humans is learned during open-ended interaction with the world. Simultaneously, open-ended learning emerged as a promising learning paradigm for training generally capable artificial agents. While open-ended learning has brought forth agents capable of solving vast amounts of diverse tasks, their intrinsic ToM capabilities are unknown. Multi-Agent environments embodying this open-ended learning paradigm are sparse, and methods to benchmark ToM capabilities are limited in their application. Thus, we extend the open-ended-learning environments XLand-MiniGrid with multiagent functionality. Using this environment we train agents via state-of-the-art autocurricula methods. We evaluate the ToM-capabilities within these agents on our handcrafted ToM Benchmark. Our experiment results reveal that our open-ended learning environment poses a challenging problem. We report evidence of lacking intrinsic ToM-capabilities within the trained agents. Our ToM benchmark thus remains an open challenge to be attempted by future work.