CrustAPI vs serper.dev benchmark
Run: July 11, 2026, from a single machine (macOS, residential US connection).
Method: identical queries sent to both APIs back to back, wall-clock time measured
around each HTTPS request, medians reported. Keys redacted. Re-run it yourself with
the script at the bottom.

== WEB SEARCH (5 queries, run once each) ==
Queries: "best crm software", "coffee roasters brooklyn", "how to start an llc",
         "running shoes for flat feet", "b2b lead generation tools"

serper   POST google.serper.dev/search
  timings (ms): 1529, 1014, 1078, 1517, 955   -> median 1078
  organic results per query: 8, 10, 8, 8, 8

crustapi GET crustapi.com/v1/search?type=web
  timings (ms): 1347, 2477, 1825, 1178, 949   -> median 1347
  organic results per query: 9, 8, 8, 7, 8

== IMAGES ("latte art", 1 query each) ==
serper   1100 ms, 10 images returned, 1 credit charged (default num)
crustapi 1657 ms, 99 images returned, 1 credit charged

serper with num=100 (same query): 100 images returned, 2 credits charged.
So at serper's best configuration: 50 images per credit. CrustAPI: ~100 per credit.

== NEWS ("artificial intelligence") ==
serper   966 ms, 10 items
crustapi 1738 ms, 10 items

== AUTOCOMPLETE ("why is the sky") ==
serper   553 ms, 10 suggestions
crustapi 286 ms, 10 suggestions
(A later re-run measured crustapi at 410 ms. Latencies vary run to run; we claim
"faster in our runs", not a fixed multiple.)

== ZERO-RESULT BILLING PROBE ==
serper   POST /search q="\"zxqv9kqpw waaxbolt mnrtq88x\"" (quoted nonsense phrase)
         -> organic: 0 results, credits charged: 1
crustapi GET /v1/search?type=web&q=... with a zero-result query
         -> charged 0 credits (billing is per successful search; verified in
            production repeatedly, including via the dashboard mini-batch)

== HONEST NOTES ==
- serper was faster than us on web and news in this run. We were faster on
  autocomplete. Both APIs live in the ~1 second band; speed should not be the
  reason anyone switches in either direction.
- Single-run numbers, one machine, one day. Network conditions matter. Run the
  script below from your own infrastructure before deciding anything.

== SCRIPT (python3, stdlib only; put your own keys in) ==

import json, time, urllib.request, statistics
from urllib.parse import urlencode

SERPER_KEY='...'; CRUST_KEY='...'
QUERIES=['best crm software','coffee roasters brooklyn','how to start an llc',
         'running shoes for flat feet','b2b lead generation tools']

def serper(endpoint, body):
    t0=time.time()
    req=urllib.request.Request(f'https://google.serper.dev/{endpoint}',
        data=json.dumps(body).encode(),
        headers={'X-API-KEY':SERPER_KEY,'Content-Type':'application/json'})
    d=json.loads(urllib.request.urlopen(req, timeout=30).read())
    return (time.time()-t0)*1000, d

def crust(params):
    t0=time.time()
    req=urllib.request.Request('https://crustapi.com/v1/search?'+urlencode(params),
        headers={'x-api-key':CRUST_KEY})
    d=json.loads(urllib.request.urlopen(req, timeout=30).read())
    return (time.time()-t0)*1000, d

s_ms=[]; c_ms=[]
for q in QUERIES:
    ms,d=serper('search',{'q':q}); s_ms.append(ms)
    ms,d=crust({'type':'web','q':q}); c_ms.append(ms)
print('serper median', statistics.median(s_ms))
print('crust  median', statistics.median(c_ms))