And how well did they do?
The abstract from the relevant Hydrological Sciences Journal paper is pretty grim.
The paper is not too long and recommended reading. Especially Fig8 which shows how clumsy the models mismatch with reality.
From the Results
This clearly shows that GCMs totally fail to represent the HK-type climate of the past 100–140 years, which is characterized by large-scale over-year fluctuations (i.e. successions of negative and positive "trends") that are very different from the monotonic trend of climatic models. In addition, they fail to reproduce the long-term changes in temperature and precipitation (Fig. 8). Remarkably, during the observation period, the 30-year temperature at Vancouver and Albany decreased by about 1.5°C, while all models produce an increase of about 0.5°C (Fig. 8, lower left). With regard to precipitation, the natural fluctuations are far beyond ranges of the modelled time series in the majority of cases (Fig. 8, lower right).
The systematically unsatisfactory agreement of modelled and observed time series can have four interpretations: (1) the models are poor; (2) the data are poor; (3) the modelled and observed time series are not comparable to each other (e.g. there should not be a direct link between observations at a point and model outputs at neighbouring grid cells); (4) our calculations and comparisons are wrong.
This bit from the conclusion is very damning:
At the annual and the climatic (30-year) scales, GCM interpolated series are irrelevant to reality. GCMs do not reproduce natural over-year fluctuations and, generally, underestimate the variance and the Hurst coefficient of the observed series. Even worse, when the GCM time series imply a Hurst coefficient greater than 0.5, this results from a monotonic trend, whereas in historical data the high values of the Hurst coefficient are a result of large-scale over-year fluctuations (i.e. successions of upward and downward "trends"). The huge negative values of coefficients of efficiency show that model predictions are much poorer than an elementary prediction based on the time average. This makes future climate projections at the examined locations not credible. Whether or not this conclusion extends to other locations requires expansion of the study, which we have planned. However, the poor GCM performance in all eight locations examined in this study allows little hope, if any. An argument that the poor performance applies merely to the point basis of our comparison, whereas aggregation at large spatial scales would show that GCM outputs are credible, is an unproved conjecture and, in our opinion, a false one. Our future plan also includes a study of this question after refinement and extension of our methodology.
In a Journal Paper saying that a method is "irrelevant to reality" is a formal way of saying it's junk, that it has no predictive value. They go on to say that a basic interpolation of existing data would give better results than a Global Climate Model would. The gauntlet has been thrown down.
This is how science is supposed to work. Models are tested to see if they have predictive value. Results and methods are reviewed by peers, colleges, and rivals.
If Koutsoyiannis, et al. are wrong then someone should come along with evidence showing how they were wrong. Just like how Koutsoyiannis, et al. presented a compelling case as to the lack of utility of Global Climate Models.
It will be interesting to the see papers generated in response.
This is not to say Global Warming is not happening or that Humans are not causing it, but this is strong evidence that models used to predict future climate are not reliable at all.