Kraken vs Gemini: which API is more reliable?
Over the 60 days of a 90-day window up24 has measured so far, Kraken was more available — 99.99% against 99.94% of checks answered, and Kraken was quicker — a typical p95 of 65ms against 433ms.
Both venues are probed by the same probes on the same schedule — 90-day window, one request every 5 seconds, percentiles over successful responses only and measured from Lauterbourg, FR. Neither exchange reports any of this; up24 measures it.
| MEASUREMENT | ||
|---|---|---|
Uptime · 90d | 99.99% | 99.94% |
REST checks behind it | 3,083,455 | 3,007,147 |
Days measured | 62 of 90 | 60 of 90 |
Typical p50 | 27ms (better) | 118ms |
Typical p95 | 65ms (better) | 433ms |
Worst day | 99.95% | 98.67% |
up24 incidents · 90d | 77 | 421 |
WS delivering · 24h | 99.86% | 99.99% (better) |
WS connected · 24h | 99.97% | 100.00% (better) |
Reconnects · 24h | 58 | 0 (better) |
Sequence gaps · 24h | not published | 0 |
Longest feed silence · 24h | 1 min 1s (better) | 3 min 52s |
Own status feed | statuspage | statuspage |
Incidents they acknowledged | 39 of 77 | 30 of 421 |
Median gap to their own notice | 1h 58m later | 1h 20m later |
KRAKEN
Over the 62 days of a 90-day window up24 has measured so far, Kraken's REST API answered 99.99% of 3,083,455 checks, with a typical p95 of 65ms and a typical p50 of 27ms. Its worst day was 14 SEP 2026, at 99.95%. up24 opened 77 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.
Over the last 24 hours Kraken's WebSocket feeds were delivering 99.86% of the time on a socket up24 held open 99.97% of it, across 58 reconnects, (sequence numbers are not checked for this feed), with a longest silence of 1 min 1s.
GEMINI
Over the 60 days of a 90-day window up24 has measured so far, Gemini's REST API answered 99.94% of 3,007,147 checks, with a typical p95 of 433ms and a typical p50 of 118ms. Its worst day was 25 SEP 2026, at 98.67%. up24 opened 421 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.
Over the last 24 hours Gemini's WebSocket feeds were delivering 99.99% of the time on a socket up24 held open 100.00% of it, across 0 reconnects, with 0 sequence gaps and a longest silence of 3 min 52s.
Uptime is ok checks over total checks, REST only. Both percentiles are medians of daily percentiles over the window rather than the window's own — percentiles do not compose, and the daily rollup is the finest grain that outlives raw retention. Green marks the better column on that row and nothing else: a venue with no measurement never wins one.