Search docs

Latency

Measured latency of every endpoint on the India .glx file.

By · Updated

Conditions

  • The India .glx file: 6,382,829 places from Overture and OpenStreetMap, 542 MB, read from local disk. No database.
  • One Mac (Apple M2 Max), release build, time measured with curl against the running server, best of 4 runs.

These are local numbers, not a hosted service level. Measured on 19 Oct 2026.

QueryTime
mumbai7 ms
pune6 ms
delhi5 ms
cafe4 ms
restaurant4 ms
hotel4 ms
pharmacy4 ms
gateway of india14 ms
india gate9 ms
mg road6 ms

Search near a point

QueryTime
cafe · Mumbai23 ms
pharmacy · Mumbai28 ms
hotel · Mumbai21 ms
cafe · New Delhi26 ms

A search near a point reads the places inside a box around the point from the spatial index before the word lists. This keeps distant places out of the answer and costs about 20 ms in a dense city.

Autocomplete

QueryTime
p3 ms
pu2 ms
pun4 ms
mum4 ms
bang5 ms
shaniwar wa10 ms
mg ro7 ms
gateway of in12 ms
caf · near Mumbai25 ms

Reverse and structured

QueryTime
reverse · Pune, 1 km2 ms
reverse · Mumbai, 5 km, 10 places3 ms
structured · FC Road, Pune11 ms

Every query is under 30 ms.

Planet

The planet .glx file: 652,242,366 places, 84 GB, same machine, best of 3, measured on 16 Oct 2026 with the file built before the 18–19 Oct parity work. The slowest queries hold common words in millions of places.

QueryTime
london1 ms
paris1 ms
eiffel tower6 ms
berlin hauptbahnhof7 ms
india gate32 ms
cafe · New York50 ms
ramen · Tokyo5 ms
new york151 ms
main street158 ms
autocomplete · new yo141 ms
autocomplete · main str172 ms
autocomplete · par2 ms
structured · Berlin, DE66 ms
reverse · Eiffel Tower1 ms
nearby · London, 1 km1 ms

Every planet query in the test set is under 200 ms. See Engines for accuracy and load against Pelias, Photon and Nominatim.