Theory of Mind (ToM) – the ability to infer the beliefs and intentions of others – is a central component of intelligent behaviour in multi-agent interaction. Recent work has explored whether artificial agents can acquire ToM-like reasoning through explicit modelling of other agents. This thesis investigates whether model-based opponent modelling (MBOM), a framework that combines recursive imagination with Bayesian belief mixing, can support such capabilities. Because MBOM was originally developed for fully observable, synchronous environments, directly applying it to the simultaneous, partially observable game Hanabi is not straightforward. This work analyses the structural incompatibilities between MBOM and Hanabi. Based on this analysis, turn-based model-based opponent modelling (TB-MBOM), a sequential adaptation of MBOM, is proposed that reformulates recursive imagination as turn-aware, multi-step rollouts over future interaction trajectories. To isolate the core reasoning mechanism from belief-estimation noise, the adapted framework is evaluated in an oracle setting with access to the true environment state during imagined rollouts. An empirical evaluation demonstrates that MBOM’s mixture weights systematically adapt to partner behaviour. While the experiments do not measure full Hanabi performance, they provide some first conceptual evidence that MBOM’s core mechanism serves as a viable pathway toward ToM-like inference in cooperative domains.