A SERP API benchmark with the rows attached: on 4 October 2026 we sent the same 40 queries to Google, Bing and DuckDuckGo through the QuanticData SERP API. 117 of 120 calls returned a full results page on the first attempt, the median call took 2.92 seconds, and 1,000 searches cost $0.50 on Bing or DuckDuckGo and $2 on Google.
Benchmarks of search APIs are usually published as rankings. This one is a measurement of one product: we tested our own API, with the method, the sample and the date stated, and the summary by engine is results-2026-10-04.csv.
What we measured
| Item | Setting |
|---|---|
| Date and time | 4 October 2026, 08:29 to 08:37 UTC |
| Endpoint | POST /v1/serp on the production API, default options, desktop, country US |
| Core set | 40 queries on each of Google, Bing and DuckDuckGo: 120 calls, one at a time per engine |
| Query mix | 10 informational, 10 commercial, 5 local, 5 navigational, 5 fresh (scores, weather, prices), 5 technical |
| Country set | 4 queries on Google from the United Kingdom, Germany, France, Brazil and Japan: 20 calls |
| Repeat set | 10 of the Google queries sent a second time a few minutes later |
| Latency | Wall-clock time of the HTTPS call from a client in Europe, in milliseconds |
| Full page | HTTP 200 with organic results in the JSON |
That is 150 calls in total. Nothing was tuned for the test: no warm-up and no cherry-picked queries. If you want the mechanics behind a call (localization, the identity layer, parsing), read how a SERP API works, end to end.
Success rate and latency by engine
| Engine | Full page, first call | Median | Organic results per page |
|---|---|---|---|
| 38 of 40 | 3.31 s | 9.9 | |
| Bing | 39 of 40 | 3.21 s | 9.8 |
| DuckDuckGo | 40 of 40 | 2.67 s | 9.8 |
| All three | 117 of 120 | 2.92 s | 1,152 in total |
The 117 full pages carried 1,152 organic results, between 9.8 and 9.9 per page on each engine, and the typical call is a three-second call. Calls are independent, so a batch runs in parallel rather than in sequence.
What each engine returns
The JSON envelope is the same for the three engines, but the blocks inside follow what each engine shows. Counts are pages that carried the block, out of the full pages of the core set.
| Block or field | Google (38) | Bing (39) | DuckDuckGo (40) |
|---|---|---|---|
| Organic: rank, title, link, display link, source, description | 38 | 39 | 40 |
| Sitelinks under a result | 24 | 11 | 0 |
| Result date | 0 | 19 | 0 |
| Highlighted words in the snippet | 0 | 0 | 38 |
| Related searches | 37 | 6 | 0 |
| AI Overview text | 21 | 0 | 0 |
| Inline videos | 0 | 21 | 0 |
| Ads | 0 | 1 | 14 |
| Knowledge panel | 0 | 0 | 3 |
| Featured snippet | 0 | 0 | 3 |
| Median HTML fetched per page | 121 KB | 517 KB | 35 KB |
Two readings matter for anyone building on this. First, Google's AI Overview arrived as text on 21 of 38 pages, and it followed intent: 17 of the 21 were on commercial and informational queries. If you track AI Overviews, sample by query type. Second, the engines disagree more than people assume: for the same query, the median overlap between Google's and Bing's first page was 3 domains out of 10, and the same 3 of 10 between Google and DuckDuckGo. One engine is one opinion.
The table lists what came back in this sample with default options. Verticals such as maps, shopping, news, images, scholar, hotels and flights are separate search types with their own blocks, and are not part of this benchmark; the Maps vertical is measured in our Google Maps test of 1,644 places.
Google from five other countries
The same four queries went to Google from the United Kingdom, Germany, France, Brazil and Japan: 18 of 20 calls returned a full page on the first call, with a median of 3.96 seconds. Four calls per country is a spot check, not a ranking of countries. The country parameter sets both the exit country and Google's locale, and location narrows it to a city.
Repeat queries come back in under a second
The 10 Google queries we sent a second time a few minutes later returned in a median of 0.98 seconds, all 10 with a full page. The response says so itself: search_metadata.from_cache is true when a recent identical query is served from the short-lived cache. For rank tracking that wants a fresh fetch every time, vary the query parameters or space the calls; for an agent that asks the same thing twice in a conversation, it is free speed.
Cost per 1,000 searches
| Engine | Price per search | Per 1,000 | Searches in the free $2 per month |
|---|---|---|---|
| Bing | $0.0005 | $0.50 | 4,000 |
| DuckDuckGo | $0.0005 | $0.50 | 4,000 |
| Google and rich verticals | $0.002 | $2.00 | 1,000 |
Billing is per successful search with no subscription and no monthly minimum, and failed requests are never billed. At list prices this whole benchmark, 150 calls, cost about $0.18. Volume tiers lower the unit price; they are on the pricing page.
The useful comparison is with your own alternative. If you only need ten blue links and do not care which engine, DuckDuckGo was the fastest and the cheapest row in this test. If you need AI Overviews, related searches or Google's ordering, that is the $2 row, and it is the one that returned them.
Limits of this benchmark
- One eight-minute window on one day, 40 queries per engine. It shows the shape, not a service-level figure.
- Latency is measured from one client in Europe to our API. Your network adds its own.
- Default options only: desktop, first page, 10 results. Deeper pages and the other verticals behave differently.
- English queries from a US exit in the core set. The country set is 4 queries per country.
- "Full page" means organic results present. We did not grade the correctness of every parsed field by hand.
- Search pages are public, but engines' terms restrict automated access and the law differs by country. Use the data responsibly.
The exact call, and what to buy
One endpoint, one key, the engine as a parameter. The same call is available as the search tool of the MCP server for agents.
curl -X POST "https://api.quanticdata.io/v1/serp" \
-H "Authorization: Bearer qd_live_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"query":"best running shoes 2026","engine":"google","country":"us"}'Set engine to google, bing or duckduckgo, add country and optionally location, and read payload.organic, payload.related_searches and payload.ai_overview. Every parameter is in the API documentation, a walkthrough with parsing examples is in how to use a SERP API, and the product page is the QuanticData SERP API: pay per successful search from $0.0005, with $2 of free API usage per month.