The IPCC comes out with major reports about every 5 to 8 years. I think the last one was in 2014 and they are aiming for 2022 ( 4 years from now) for the next one. One consistent feature of these reports is that all the models fit the past beautifully. This is not a surprise because data from the past is used to compute the values of the adjustable parameters in the models. By the time the reports circulate they are only a little bit off. They get increasingly far off from current data as time wears on. By the time the 2022 report comes out the predictions of 2022 from the 2014 report will be far off, perhaps comically so. Nevermind though. By constantly updating the data fed into the models they can make the new output somewhat resemble actual data from direct observation. But the further one goes back in time the worse the projections of current observations become. In fact , though there has been many changes in the models since the earliest mistakes made by Hanson's lab at GISS, there's been little demonstrable improvement in the accuracy of the models predictions. On the day their predictions are announced they all predict the global temperature about as well as the global observations do. Then it's downhill from there until it is time for the next report to come out, whence the models are pretty good once again for a couple years. By constantly revising the data fed into the climate simulation models they are kept close enough to the observed data to avoid ridicule, however once the data input stops being refreshed their output moves,over time, further and further away from truth. What this is telling us is that there are fundamental flaws in these models.
The fundamental flaw is well understood by experts in chaotic systems, but this never seems to get transmitted to the climate modeling community.