Climate models vs. observations

This is Zeke Hausfather's analysis, kept up to date. In December 2025 the climate scientist Zeke Hausfather (Berkeley Earth, Carbon Brief, lead author for the IPCC's seventh assessment) compared the last three generations of climate models with observed temperatures in “Comparing climate models with observations” on The Climate Brink. This page rebuilds the first five figures of that post and refreshes them every day with the latest data.

What you're looking at. Every few years the world's modelling groups run a shared set of experiments (CMIP). The historical part of each run is driven by known greenhouse gases, aerosols, volcanoes and solar changes up to a cut-off year (2000 for CMIP3, 2005 for CMIP5, 2014 for CMIP6); after that the models follow a middle-of-the-road scenario. Everything after the cut-off is a genuine projection that observations can now be checked against.

The shaded band is the spread of individual models (5th to 95th percentile), the blue line their average. Real-world temperatures wiggle more than the model average because of El Niño and La Niña; they should wander inside the band, not hug the line.

Method

Replicates figures 1–5 of Hausfather's post, following the method of his dashboard.
Models: global surface air temperature from KNMI Climate Explorer, one run per model, each counted equally. Four duplicate CMIP6 runs in KNMI's file are counted once.
Everything is shown relative to 1850–1900.
CMIP6 without ‘hot models’ keeps the 22 models whose transient climate response lies within the IPCC's likely range (1.4–2.2 °C), using the values from Hausfather et al. (2022). This also drops two models that warm too little.
Observations: Berkeley Earth, HadCRUT5, NASA GISTEMP, NOAAGlobalTemp and ERA5. Download the data (JSON).

References

Hausfather, Z. (2025). Comparing climate models with observations. The Climate Brink, 6 December 2025.
Hausfather, Z. The Climate Brink dashboard, “Models vs. Observations” tab; github.com/hausfath/climate-dashboard.
Hausfather, Z., Drake, H. F., Abbott, T. & Schmidt, G. A. (2020). Evaluating the performance of past climate model projections. Geophysical Research Letters, 47, e2019GL085378.
Hausfather, Z., Marvel, K., Schmidt, G. A., Nielsen-Gammon, J. W. & Zelinka, M. (2022). Climate simulations: recognize the ‘hot model’ problem. Nature, 605, 26–29.
Cowtan, K., Hausfather, Z., Hawkins, E. et al. (2015). Robust comparison of climate models with observations using blended land air and ocean sea surface temperatures. Geophysical Research Letters, 42, 6526–6534.
Hausfather, Z., Marvel, K., Schmidt, G. A., Nielsen-Gammon, J. W. & Zelinka, M. (2022). Supporting information for “Climate simulations: recognize the ‘hot model’ problem” [data set]. Zenodo. CC BY 4.0.
KNMI Climate Explorer, climexp.knmi.nl (CMIP3, CMIP5 and CMIP6 global means).

Have climate models got global warming right?

Method

References