In Model-Based Testing (MBT), long traces to errors complicate bug localisation. Shrinking of these long traces, aimed at finding smaller counterexamples from existing ones, traditionally operates at the trace-level. However, leveraging the detailed behavior captured by MBT models suggests potential for more effective shrinking methods. We propose a model-centric shrinking approach, the \textit{ShortestPathShrinker}. It centers around the idea that finding shorter traces in MBT is similar to finding shorter paths in graphs. Our experimental results show that the algorithm yields smaller or comparable traces to existing methods, in a faster fashion depending on the type of bug. Future work is discussed.