ClockOrbit

Reference

Leap second history

Every leap second inserted into UTC since the modern definition began on 1 January 1972, with the cumulative TAI−UTC offset and IERS Bulletin C reference where it can be cited.

Leap seconds inserted

27

Current TAI−UTC

37s

Most recent

December 31, 2016

Next

Retired by 2035 (CGPM 2022)

27 leap seconds since 1972No insertions since 2016-12-31CGPM 2022:leap secondsretired by ~2035197519801985199019952000200520102015202020252030203520401972-06-30 (TAI-UTC = 11)1972-12-31 (TAI-UTC = 12)1973-12-31 (TAI-UTC = 13)1974-12-31 (TAI-UTC = 14)1975-12-31 (TAI-UTC = 15)1976-12-31 (TAI-UTC = 16)1977-12-31 (TAI-UTC = 17)1978-12-31 (TAI-UTC = 18)1979-12-31 (TAI-UTC = 19)1981-06-30 (TAI-UTC = 20)1982-06-30 (TAI-UTC = 21)1983-06-30 (TAI-UTC = 22)1985-06-30 (TAI-UTC = 23)1987-12-31 (TAI-UTC = 24)1989-12-31 (TAI-UTC = 25)1990-12-31 (TAI-UTC = 26)1992-06-30 (TAI-UTC = 27)1993-06-30 (TAI-UTC = 28)1994-06-30 (TAI-UTC = 29)1995-12-31 (TAI-UTC = 30)1997-06-30 (TAI-UTC = 31)1998-12-31 (TAI-UTC = 32)2005-12-31 (TAI-UTC = 33)2008-12-31 (TAI-UTC = 34)2012-06-30 (TAI-UTC = 35)2015-06-30 (TAI-UTC = 36)2016-12-31 (TAI-UTC = 37)11 insertions in 10 years9 insertions in 17 years
DateTAI−UTC after
1972-01-01 (baseline)10
1972-06-3011
1972-12-3112
1973-12-3113
1974-12-3114
1975-12-3115
1976-12-3116
1977-12-3117
1978-12-3118
1979-12-3119
1981-06-3020
1982-06-3021
1983-06-3022
1985-06-3023
1987-12-3124
1989-12-3125
1990-12-3126
1992-06-3027
1993-06-3028
1994-06-3029
1995-12-3130
1997-06-3031
1998-12-3132
2005-12-3133
2008-12-3134
2012-06-3035
2015-06-3036
2016-12-3137

What a leap second is

UTC, the civil time standard the world runs on, is generated from atomic clocks. A second of UTC is exactly 9,192,631,770 oscillations of a caesium-133 atom and never varies. UT1, by contrast, is tied to Earth’s actual rotation, which is irregular. To keep UTC within 0.9 seconds of UT1, the International Earth Rotation and Reference Systems Service (IERS) occasionally inserts a one-second pause: a minute that runs from 23:59:59 to 23:59:60 to 00:00:00. That pause is a leap second. Every insertion since 1972 has been positive; a negative leap second has never been required.

Why insertions slowed after the early 1980s

Earth’s rotation is not decelerating uniformly. Tidal friction from the Moon does slow Earth on the millennial scale, but on decadal scales the spin wobbles. Coupling between the liquid outer core and the mantle, redistribution of mass after large earthquakes, and changes to the global mass distribution from glacial melt and atmospheric circulation all nudge the length of day up or down by a millisecond or two. The 1970s ran fast on insertions because the decadal trend ran one way; the 2000s ran slow because it ran the other way.

Why nothing has been inserted since 2016

Since around 2016 Earth has actually been rotating slightly faster than the long-term average. The length of day has dipped below 86,400 SI seconds on several occasions, and discussion has shifted from "when is the next positive leap second" to "do we need a negative one." A negative leap second would skip from 23:59:58 to 00:00:00 with no 23:59:59 at all. It was seriously considered for around 2026, but never scheduled, partly because operators of critical infrastructure dread the untested code path.

The 2022 CGPM decision: retire leap seconds by 2035

In November 2022 the General Conference on Weights and Measures (CGPM), at its 27th meeting, passed Resolution 4. It instructs the BIPM, in consultation with the IERS, ITU, and other relevant bodies, to propose by 2035 a new maximum value for UT1−UTC that is large enough that no further leap-second insertions would be needed "for at least a century." The widely-discussed candidate value is around 100 seconds. The exact mechanism, who maintains the new offset, and how it is published are still being worked out, but the policy direction is set: routine leap seconds are on their way out.

Software pain: why operators hate leap seconds

Leap seconds have repeatedly broken software that assumed a minute always has sixty seconds. The 2012 insertion exposed a livelock in the Linux kernel’s hrtimer subsystem that pegged CPUs at 100% across thousands of machines; sites like Reddit, Mozilla, and LinkedIn went down or degraded for hours. The 2015 insertion caused renewed trouble for Java applications and a number of consumer services. Database engines have tripped on the same minute name appearing twice in a row of timestamps.

Large operators responded by smearing the leap second across a longer interval. Google’s public NTP servers and Amazon Time Sync stretch the extra second across a 24-hour window so that, from the perspective of any client, the clock simply runs a tiny bit slow for a day rather than visibly jumping. Smearing is incompatible with strict UTC, but it has proved far more compatible with software written by humans.

Related references

For why a star-relative day differs from a solar day, see the sidereal day clock. For other unusual units of time, including the SI second’s formal definition, see the time units reference.