Fun fact: most modern forecasts from individual centres are combinations of numerous forecasts with slightly different start conditions (perturbations) which are called ensembles.
No one model is correct everywhere at every lead time. ECMWF is very strong 5-10 days, MO is better 3-5. But it depends on how you measure it and where in the world you measure it.
Talking about ocean temperatures a lot of the global models are now coupled with corresponding ocean models so you get a more complete modelling of the physics impacting both (somewhat artificially separated) systems.
The documentation here is a pretty good started talking about how forecasts are made and explaining some of the variability.