Mesoscale eddies are prevalent in the global oceans and are known to influence oceanic and atmospheric conditions. This study aims to assess how the impact of mesoscale eddies on air-sea CO2 fluxes varies throughout their lifecycle. We compared six machine learning models, including light gradient boosting machine, support vector machine, and XGBoost models, to construct an FCO2 evaluation model. Among these, the XGBoost model performed the best, with model validation yielding the coefficient of determination (R-2) value of 0.9046 and a root mean square error of 1.4165 mmol m(-2) d(-1), successfully assessing the air-sea CO2 fluxes in the South Atlantic. Analysing eddies with a lifespan exceeding 300 d during the period from 1995 to 2020, we identified two distinct peaks in the influence of eddies on air-sea CO2 fluxes during their life cycle: the first peak occurs approximately between the second and fifth deciles of the life cycle, during which the eddies have a broader impact range but weaker influence; the second peak emerges between the seventh and ninth deciles, exhibiting a narrower impact range but significantly stronger influence. These findings provide crucial quantitative evidence for understanding the marine carbon sequestration mechanism and reveal the complexity and dynamic nature of mesoscale eddies' impact on air-sea CO2 fluxes.