Abstract Real‐time forecasts of global mean surface temperature have been issued annually for 2000–2025. The 25 forecasts for 2000–2024 have had high skill. Correlations and root mean square errors (rmse) between observations and real time forecasts, statistical forecasts, and dynamical forecasts (the latter from 2008 only) were 0.96°C, 0.95°C, 0.94°C and 0.09°C, 0.11°C, 0.10°C, respectively. Interannual forecast correlations independent of the trend are also high and statistically significant at 0.73, 0.76, and 0.61, respectively. High correlations reflect the fact that the observed strong rate of warming interannual variability was well captured. Finally, we discuss the 2023 and 2024 forecasts. The extreme and widely unexpected jump in global warming in 2023 was not well captured, but the record warmth of 2024 was better forecast, especially by the physics‐based dynamical model.