Above all, climate models must reflect the real-world climate with acceptable accuracy. So, models are first run using known atmospheric conditions and then validated against observed climate. Validation might include, for example, an assessment of the ability of the model to recreate known features of observed climate, and examination of how well extreme events, such as heatwaves, are represented by the model.
Assuming the physics of the validated model to be consistent for all situations, it can be used to simulate weather over the next season or climate in the future, such as might be expected under Global Warming. Also, the model output can be used to drive impact models of things like agriculture, health, forest fires, etc. that directly assess the likely effects of seasonal variability and climate change.
One of the most important scientific developments in recent years is the move from single modelling centres running and using their own model towards a more co-operative approach in which scientists work with outputs from many different centres. This multimodel or ensemble approach is the one taken in the ENSEMBLES project which involves more than 70 research institutions.