Reliably fast, not just fast once
Edge time-to-first-byte over the last 30 days. The bold line is the median across regions across 7 regions; the shaded band is fastest-to-slowest — that band is the “fast from everywhere” claim, drawn. Each region is a faint line of its own, so the aggregate is never the only thing on screen. The probed region set changed over this window (6–7 regions); the aggregate is labelled with its region count so a shift from a region dropping in or out is never read as a change in speed.
A missing run is a gap in the line, never a straight line drawn
through it — we read the series off without interpolation, the same rule
/v1/curve follows. The full series is open at
/v1/latency.
batchwatch runs at Cloudflare's edge, so calling us does not add a round trip to a distant origin: the answer comes from close to wherever your pipeline runs. From every region we measured, the edge starts replying in 159–461 ms. Measured 49 minutes ago.
| Region | Probe | Edge TTFB | Total |
|---|---|---|---|
| Western Europe | Falkenstein, DE | 159 ms | 207 ms |
| Northern America | Buffalo, US | 251 ms | 329 ms |
| Southern Africa | Johannesburg, ZA | 283 ms | 461 ms |
| South America | Sao Paulo, BR | 306 ms | 528 ms |
| Southern Asia | Mumbai, IN | 387 ms | 411 ms |
| Australia and New Zealand | Sydney, AU | 454 ms | 479 ms |
| Eastern Asia | Tokyo, JP | 461 ms | 488 ms |
Time to first byte from a real GET batchwatch.dev/v1/wait?provider=openai&model=gpt-5.6-sol — the
same open API call an integrator makes, running the real database query, not
a static health check. The number therefore includes our query time. Measured
from the free Globalping
probe network; regions with no usable probe are left out rather than
estimated.