Gemini WS · connection degradation, 3 min on 18 SEP 2026
Gemini WS — 6.7% of windows failing over the last five minutes (conn) — from Singapore, SG and Carlstadt, US. up24 opened this incident from its own measurements of Gemini's public API on 18 SEP 2026 07:25 UTC. It lasted 3 min and ended 18 SEP 2026 07:28 UTC, when the component started answering again.
WHAT HAPPENED
- up24 opened the incident at degraded, conn.
- The component started answering again. up24 dates the incident from this moment, not from when its clearing rule finished.
HOW IT FAILED
Stream measurements are kept for seven days, and this incident is older than that. The incident itself is permanent; the ten-second windows behind it are not.
A stream has no latency of its own — a feed either delivers or it does not — so this is Gemini's REST endpoints over the same window, per minute, as the median across them. It answers the question the timeline cannot: whether the rest of the exchange was fine while this feed was down.
DID GEMINI SAY SO?
Gemini publishes a status feed and nothing in it lines up with this incident. That is not proof they stayed silent; it is what their feed says.
up24 opened 421 incidents for Gemini in the last 90 days; 30 fall within six hours of something Gemini published, and the rest were never mentioned. At the median, Gemini's own notice came 1h 20m before up24 opened its incident. Pairing is by time alone — nothing checks that a notice is about the fault up24 measured, so the 11 that up24 timestamped first are not scoops. Every entry links to the venue's own notice so the pairing can be judged.
GEMINI — OTHER INCIDENTS
Measured by up24's own probe against Gemini's public API, btcusd against USD, every 5s. This page is generated from the incident's own rows — 2 recorded transitions — and nothing on it is reported by Gemini.