Widespread evidence of a +4-6-m sea-level highstand during the last interglacial period (Marine Isotope Stage 5e) has led to warnings that modern ice sheets will deteriorate owing to global warming and initiate a rise of similar magnitude by ad 2100 (ref. 1). The rate of this projected rise is based on ice-sheet melting simulations and downplays discoveries of more rapid ice loss. Knowing the rate at which sea level reached its highstand during the last interglacial period is fundamental in assessing if such rapid ice-loss processes could lead to future catastrophic sea-level rise. The best direct record of sea level during this highstand comes from well-dated fossil reefs in stable areas. However, this record lacks both reef-crest development up to the full highstand elevation, as inferred from widespread intertidal indicators at +6 m, and a detailed chronology, owing to the difficulty of replicating U-series ages on submillennial timescales. Here we present a complete reef-crest sequence for the last interglacial highstand and its U-series chronology from the stable northeast Yucatán peninsula, Mexico. We find that reef development during the highstand was punctuated by reef-crest demise at +3 m and back-stepping to +6 m. The abrupt demise of the lower-reef crest, but continuous accretion between the lower-lagoonal unit and the upper-reef crest, allows us to infer that this back-stepping occurred on an ecological timescale and was triggered by a 2-3-m jump in sea level. Using strictly reliable (230)Th ages of corals from the upper-reef crest, and improved stratigraphic screening of coral ages from other stable sites, we constrain this jump to have occurred approximately 121 kyr ago and conclude that it supports an episode of ice-sheet instability during the terminal phase of the last interglacial period.