Abstract. Even the most advanced climate models struggle to reproduce the observed wintertime circulation of the atmosphere over the North Atlantic and Western Europe. During winter, this particularly challenging region is dominated by eddy-driven and highly non-linear flows, which are often studied from the perspective of regimes – a small number of qualitatively distinct atmospheric states. Poor representation of regimes associated with persistent atmospheric blocking events, or variations in jet latitude, degrade the ability of models to correctly simulate extreme events. In this paper we leverage a recently developed hybrid approach – which combines both jet and geopotential height data – to assess the representation of regimes in 8,400 years of historical climate simulations drawn from CMIP6, CMIP5 and HighResMip. We show that these geopotential-jet regimes are particularly suited to the analysis of climate data, with considerable reductions in sampling variability compared to classical regime approaches. We find that CMIP6 has a considerably improved spatial regime structure, and a more trimodal eddy-driven jet, relative to CMIP5, but still struggles with underpersistent regimes, and too little European blocking, when compared to reanalysis. Reduced regime persistence can be understood, at least in part, as a result of jets that are too fast and eddy feedbacks on the jet stream that are too weak – structural errors that do not noticeably improve in higher resolution models.