It was the vision of Jens Schröter that variable-resolution (unstructured) meshes can be useful in modeling of large-scale flows because of their geometric flexibility, already exploited in modeling of coastal, tidally-driven dynamics. This vision materialized in the development of FESOM, the unstructured-mesh sea-ice ocean model. Originally the model relied on the finite-element method and used linear continuous tetrahedral elements. The early version of the model was constrained to the North Atlantic. This version, described in Danilov et al. (2004), followed the finite-element wisdom: it used implicit time-stepping and relied on iterative solvers and finite-element stabilisation techniques. This approach proved to be too inefficient numerically and was abandoned in favor of an approach where only the sea surface height evolution is treated with the implicit method which is followed since then. The basic principles of new dynamical core were presented in Wang et al. (2008). The name of FESOM was coined by Timmermann et al. (2009), who implemented coupling between the ocean model and the finite-element sea ice model (FESIM) and performed first simulations of global ocean circulations thus paving the way to realistic FESOM applications. FESOM was getting mature through its participation in CORE-II intercomparison project (see virtual special issue of Ocean Modelling), and the effort on tuning FESOM (version 1.4) is documented in Qiang et al. (2014). It was coupled to the atmospheric model ECHAM6, as described by Sidorenko et al. (2015), forming the AWI Climate model (AWI-CM) participating in CMIP6. On the road to FESOM 1.4 we experimented with tetrahedral and prismatic elements (Wang et al. 2008), staying in the end with the former to ensure general vertical levels. We also experimented with non-conforming linear elements, keeping the conforming elements as most economical. While FESOM1.4 stays at the heart of numerous practical applications, the development is focused on FESOM2.0 with finite-volume dynamical core which ensures much higher numerical throughput through much more efficient memory access and infrastructure. This version is documented in Danilov et al. (2017). The ingredients of FESIM are summarized in Danilov et al. (2015).