We compare the short- and medium-range predictability of weather regimes of a quasigeostrophic model as defined by a hierarchical cluster algorithm and a Lyapunov-based clustering method recently introduced in the literature. Both procedures lead to weather regimes displaying very different predictability properties on the short and medium range bases. While the former does not distinguish between stable and unstable weather regimes, the latter leads to clusters which do not display a good medium range predictability. We introduce a new clustering method taking advantages of the two previous techniques. Its application in the context of the quasi-geostrophic model gives rise to regimes possessing at the same time a good medium range skill and well separated instability properties, indicating the possibility to build a systematic cartography of the short-term predictability of weather fields in phase space.