Here are surface temperatures compared to CMIP5 models:
"CMIP" stands for Climate Model Intercomparison Project. It's a database containing results from climate models for many different scenarios, that climate modelers use to compare their results to each other and to observations. It's now on its 5th version. I talked to a climate modeler a while back, and she said modelers are usually more interested in how the models compare than how they match reality. The models all make different assumptions, so learning which model best forecast, say, the Atlantic Meridional Overturning Circulation, can give good clues. Or modelers might replace one component of their model to see how it compared to the earlier version.
Clearly, climate models are doing a pretty good job. At least for global mean surface temperature.
The model results for years prior to 2017 use "historical" forcings -- the actual forcings that occurred then, based on the measured values of greenhouse gases, etc. That's only fair -- it would make no sense to compare a 1990 model with a scenario that didn't happen. After all, no climate model can foresee the future of GHG emissions, volcanic eruptions, and changes in solar irradiance.
And here are the results for HadCRUT alone: