The global carbon budget imbalance encapsulates the inaccuracies in state-of-the-art estimates of anthropogenic carbon dioxide emissions, land and ocean uptake, and atmospheric growth rate estimates. Growth rates derived from marine boundary layer concentration observations are assumed to be highly accurate, while land and ocean process models are considered less accurate, often being cited as the primary cause of the imbalance. Here we show substantial discrepancies in these growth rate estimates when representing the entire atmosphere. This discrepancy accounts for 25% of the 0.76 petagrams per year root-mean-square imbalance reported in the 2023 Global Carbon Project (GCP) report. Further investigation into the imbalance metric across seven annual GCP reports reveals quantitative evidence of improving process models and inventory emissions estimates due to enhanced forcing data, larger model ensembles, and the inclusion of new processes. Overall, we report a 37% reduction in the imbalance, from 0.91 to 0.57 petagrams per year, between the 2017 and 2023 GCP reports by combining process model and inventory improvements with atmospheric growth rate corrections. Our study indicates that land and ocean process models are more accurate than previously assumed, and that the scientific understanding of Earth’s carbon cycle is improving.