Seasonal predictability is an active field of research given its strong potential to guide decision-making in many societal and economic sectors. In this study, we compare the predictive skill of the climate model EC-Earth3 at two different horizontal resolutions. The standard resolution – SR – (high resolution – HR) is of around 70 (40) km in the atmosphere and 100 (25) km in the ocean. Both forecast systems are initialised in the same way in May and cover the period 1990–2015, with a forecast period of 8 months. We focus on the Tropics, and particularly on El Nino Southern Oscillation (ENSO), the main source of predictability at seasonal timescales. Statistically significant improvements are found in HR with respect to SR for predicting ENSO. However, the predictive skill drops quickly in the Western Equatorial Pacific (WEP) in both configurations, more pronouncedly in SR. The poor skill in the WEP is directly linked to a misrepresentation of its relationship with the ENSO region, which is ultimately associated with an overly strong westward extension of ENSO-related variability, a model error more pronounced in SR. This erroneous spatial simulation of ENSO is related to the mean cold bias of the cold tongue, which progressively extends westwards with the forecast time. We show that an overly weak air-sea coupling, more pronounced in SR, prevents the model from simulating the correct ENSO development. We also show that a better simulation of the Atlantic Nino teleconnection with the tropical Pacific in HR compared to SR leads to better ENSO prediction. Improving model resolution can increase the predictive skill of forecast systems by improving the simulation of the mean state and atmospheric teleconnections. However, ENSO simulation errors and mean state biases need to be better understood to improve forecasts, in particular in the WEP, a region of convection particularly important for teleconnections to extratropics.