Expectations are generated with different degrees of predictive uncertainty prior to onset of musical events. This study explored influences of genre specific expertise in non-musicians, classical, and jazz musicians listening to unfamiliar Charlie Parker solos. Two probabilistic computational models of expectation were trained: one on folksongs (General), the other on jazz (Bebop). Twenty-four melodies were selected whose final notes differed in Shannon entropy estimated by the two models. Listeners' uncertainty was assessed explicitly and inferred from expectedness ratings of different continuation tones. The analysis showed that jazz musicians followed 'Bebop' and non-musicians followed 'General'. Classical musicians showed some decoding of the jazz style, utilising a somewhat underdeveloped version of 'Bebop'. Moreover, experts experienced more salient prediction errors in low-entropy contexts, and musical skills predicted the extent of cognitive model optimisation. Our results suggest that expertise entails both possessing an accurate predictive model and selecting an optimal model for the given context.