How accurate have computer global warming models been so far? Is there consistency between models?

The main problem with modeling any type of behavior involves a very common base assumption, that things will go on as they have in the past.  Sometimes that means a trend will continue, or sometimes that cyclic behavior will continue.  The largest of most long term changes, however, are often caused by the catastrophic or unpredictable, something that isn't foreseen, and thus can't be built into a model. 

It is easy to build a model that suggests a trend will continue indefinitely, or that it will continue with minor cyclic variations.  It is also easy to build a model that allows for a one-in-a-million event.  However, it is really hard to "sell" that model.  You would be saying, in effect, my model predicts "outcome A"  99.9999 percent of the time, but we can instead expect outcome B to actually occur, even though the model barely allows for it. 

Unfortunately for models, and those who love them, a lot of big changes involve those rare events.