Unix ns / μs

This page converts POSIX time at four resolutions: seconds, milliseconds, microseconds, and nanoseconds. Paste an integer and leave Unit on Auto, or force a unit when the digit length is misleading. Pick a UTC calendar date to emit all four integers at midnight. The everyday Unix timestamp converter still owns live “now” and common second-or-millisecond debugging. This spoke is the BigInt round-trip for μs and ns, so low digits are not rounded away before the civil ISO line is shown.

Digit length is the unit checksum

Languages disagree on the unit hiding inside an integer named epoch. Go’s time.UnixNano, Java’s System.currentTimeMillis, Python’s time.time_ns, MySQL FROM_UNIXTIME, and JavaScript Date.now() all sit on the same timeline with different multipliers. A ten-digit value is usually whole seconds. Thirteen digits are usually milliseconds. Sixteen digits are usually microseconds. Nineteen digits are usually nanoseconds. Pasting a nineteen-digit payload into a seconds field yields a civil date in the far future and a silent production bug.

Auto-detect on this page uses the length of the absolute digit string: up to twelve digits as seconds, thirteen or fourteen as milliseconds, fifteen or sixteen as microseconds, and seventeen or more as nanoseconds. Override the Unit menu when a clock stores leftover trailing zeros, or when a test fixture pads a millisecond to nineteen digits. Digit length is a checksum, not a proof; a known civil midnight in 2026 is the better proof.

The sibling batch timestamp converter is for many second or millisecond lines. Hexadecimal seconds belong on the hex Unix page. Start and end of a UTC day belong on epoch bounds, which prints range edges rather than a single mystery integer.

Why BigInt instead of JavaScript Number

JavaScript’s Number cannot hold every nanosecond since 1970 as an integer. MAX_SAFE_INTEGER is 2^53−1. A nineteen-digit nanosecond already exceeds that safe integer range, so ordinary division rounds the low digits before you ever format a date. This page uses BigInt for microsecond and nanosecond input so the integer is divided with exact integer arithmetic, then converted to milliseconds only for the Date display step.

That display step still snaps to JavaScript Date millisecond resolution. Listed microseconds and nanoseconds in the result are reconstructed from the millisecond instant (ms×1,000 and ms×1,000,000). Sub-millisecond remainder from the original integer is not preserved after that reconstruction. Keep the original BigInt in the log if a forensic remainder below one millisecond matters.

Negative timestamps before 1970 are valid POSIX values and are accepted. Very large future values may be outside the range Date can represent; the form errors instead of wrapping. This is a browser clock limit, not a POSIX spec limit.

How to convert without a hardcoded civil example

Paste the integer from the log. Leave Unit on Auto unless you already know the column’s unit. Read the ISO-8601 UTC line, then read the four listed integers. To go the other way, pick a calendar date; the page emits seconds, milliseconds, microseconds, and nanoseconds for that UTC midnight. Recalculate a sample for 2026 when writing docs so screenshots match current digit lengths.

A Python service that stores time.time_ns() and a browser chart that uses Date.now() must meet on the same instant. Convert one sample both ways so the chart’s millisecond axis matches the nanosecond column after dividing by one million. Database migrations from DATETIME to BIGINT should name the unit in the column (ts_ms versus ts_ns). Convert a known civil date here and paste the integer into a unit test.

Share seconds in URLs on the classic Unix page. Use this page when the payload is microseconds or nanoseconds. Sort keys in data lakes should stay integers, not formatted strings. Convert a known midnight in 2026 once, store the integer, and never parse locale dates back from CSV exports.

POSIX days, leap seconds, and other epochs

POSIX time does not include leap seconds. A nanosecond field cannot represent 23:59:60. Civil days here have 86,400 labeled seconds. For TAI or GPS linear seconds, which do not skip leap seconds the same way, use GPS / TAI time. Do not expect this converter to prove whether a leap second was inserted.

This is not a wall-clock timezone converter. Attach a zone only when displaying; store UTC integers. Wrong epoch plus wrong unit stacks two errors. If the integer might actually be FILETIME, WebKit, Cocoa, or NTP, start from the converters hub instead of forcing a Unix unit.

.NET 100-nanosecond ticks count from year 1, not from 1970. Those values belong on the .NET ticks converter even though the tick length is in the same ballpark as this page’s nanosecond field. A 18- or 19-digit integer is not automatically Unix nanoseconds.

Schema comments for 2026

If a column is named created_at but holds nineteen digits, assume nanoseconds until proven otherwise; if it holds thirteen, assume milliseconds. Write that guess in the schema comment the same day you convert a sample here. APIs should label the unit in the field name or in the OpenAPI description so clients do not sniff digit length in production.

Label the unit in logs too. A line that prints only “epoch=…” without s, ms, μs, or ns forces every reader to guess. Recalculate examples for 2026 when the runbook is reprinted so nobody copies a screenshot whose digit length no longer matches the code.

If two services disagree by a factor of 1,000, they are on adjacent units, not on different epochs. If they disagree by a factor of 1,000,000, one side is nanoseconds and the other is milliseconds. Convert one shared midnight through this page and paste both integers into the failing test. Do not “fix” the civil date string in a CSV; fix the unit in the writer.