Most CAMS and C3S data data is produced and archived not on a Cartesian grid (a two-dimensional rectangular grid), but on a reduced Gaussian grid - think of it as a globe with a series of evenly spaced data points along each parallel, and parallels spaced at regular intervals. So near the poles you have only few points along a parallel, but close to the equator you have about many data points along a parallel. For example for ERA-Interim non-wave data this interval is about 0.75 degrees[1], and for ERA5 the default resolution the interval is about 0.3 degrees.
When you download CAMS and C3S data, you can specify to get the output data in a Cartesian lat/long grid and at a custom resolution.
For ERA-Interim you can specify a custom grid on the data server. In this case the default horizontal resolution is 0.75x0.75 degrees (about 80km), approximating the irregular grid spacing on the native Gaussian grid.
With For ERA5 you can specify a custom grid using a Python script. In this case you should set the horizontal resolution to maximum 0.3x0.3 degrees (about 30km), approximating the irregular grid spacing on the native Gaussian grid.
In principle you can specify a higher output resolution, but the system does not use more input points, it merely interpolates the same data to a higher resolution. This makes the output look smoother, but does not increase the accuracy or the precision of the data.
When data is interpolated, all continuous fields (e.g. precipitation, temperature) are interpolated by bilinear interpolation, and discrete fields (e.g. vegetation, precipitation type, soil type) and Wave 2D spectra are interpolated by nearest-neighbour. For more information about our grids and interpolations see in this presentation https://software.ecmwf.int/wiki/download/attachments/55122669/intro-interpolation-2016.pdf?api=v2
...