Binance API status

Over the 62 days of a 90-day window up24 has measured so far, Binance's REST API answered 99.99% of 3,131,254 checks, with a typical p95 of 402ms and a typical p50 of 158ms. Its worst day was 21 AUG 2026, at 99.99%. up24 opened 38 incidents against it in that window — uptime counts REST checks only, and an incident is as often a WebSocket feed that stalled or a socket that dropped. Is Binance down right now? →

No successful measurements in this range yet.

Latency is measured from each region separately and never averaged between them — the distance is the measurement. Uptime, incidents and the 90-day bars above are the opposite: they combine every region by quorum, so one probe box losing its route to an exchange is not recorded as that exchange going down. Quorum is a counting rule and it only holds while a majority can still see the venue; where a vantage point cannot measure this venue at all, its refusals are named below and left out of every number on the page rather than being left to a vote.

US East — not measured. Binance does not serve Carlstadt: 51,804 checks were refused there in the last day, since 30 Aug 2026, and none of them is a measurement of whether Binance is up. They are out of this row's uptime and out of the 90-day figure. Carlstadt is refused with HTTP 451, Unavailable For Legal Reasons on every REST endpoint, on every check, while Lauterbourg and Singapore are served normally. The venue says why in the body of the refusal: “Service unavailable from a restricted location according to ‘b. Eligibility’ in https://www.bin…” — the venue names the clause it refuses under, and the URL is cut off mid-host because the capture kept 120 bytes on the day it was taken. A refusal from the venue’s own edge, not a route that failed, and not a measurement of whether the venue is serving anybody else — so a reading carrying this class is discounted wherever it was recorded, the socket’s ten-second windows included — though the refused count beside this note is REST checks only.

38 incidents over 62 measured days, grouped into 8 distinct faults. A component that fails the same way twice a week has one fault many times, not many faults, and a list that prints it once per firing hides the difference behind its own length.

WORST · 90D

BY FAULTranked by how often each fired · every occurrence links to its own page

Incidents are opened by up24's own state machine over its own measurements, not by Binance's status page. One WebSocket carries every stream a venue publishes, so a dropped socket is recorded once, against the connection, rather than once per stream.