To determine the most important drivers of successful ageing at extreme old age, we combined community-based prospective cohorts: Tokyo Oldest Old Survey on Total Health (TOOTH), Tokyo Centenarians Study (TCS) and Japanese Semi-Supercentenarians Study (JSS) comprising 1554 individuals including 684 centenarians and (semi-)supercentenarians, 167 pairs of centenarian offspring and spouses, and 536 community-living very old (85 to 99 years). We combined z scores from multiple biomarkers to describe haematopoiesis, inflammation, lipid and glucose metabolism, liver function, renal function, and cellular senescence domains. In Cox proportional hazard models, inflammation predicted all-cause mortality with hazard ratios (95% CI) 1.89 (1.21 to 2.95) and 1.36 (1.05 to 1.78) in the very old and (semi-)supercentenarians, respectively. In linear forward stepwise models, inflammation predicted capability (10.8% variance explained) and cognition (8(.)6% variance explained) in (semi-)supercentenarians better than chronologic age or gender. The inflammation score was also lower in centenarian offspring compared to age-matched controls with Δ (95% CI) = - 0.795 (- 1.436 to - 0.154). Centenarians and their offspring were able to maintain long telomeres, but telomere length was not a predictor of successful ageing in centenarians and semi-supercentenarians. We conclude that inflammation is an important malleable driver of ageing up to extreme old age in humans.