Scrapeless is a Google Search API built as a direct Serper.dev alternative. Serper publishes response times of 1–2 seconds; Scrapeless publishes under 1 second.
Both start around $1 per 1,000. The difference is what happens after that: Scrapeless bills per successful request with no prepaid pack and no credits that expire, charges the same for result 11 as for result 1, and ships a dedicated Google AI Overview endpoint.
Serper.dev
1–2s
Stated response time
Taken from each company’s own pricing and product pages, October 2026.
Scrapeless
<1s
Stated response time
Taken from each company’s own pricing and product pages, October 2026.
Serper's list rate at volume is genuinely low — lower per credit than our base rate, and we are not going to pretend otherwise. What the headline rate leaves out is the cash you commit up front, the six-month expiry on that credit, and the second credit charged once you page past result 10. Those three things decide your real cost per usable row.
| Credits | Scrapeless | Serper pack | Upfront | Per 1,000 credits | Expires |
|---|---|---|---|---|---|
| 2,500 | Free trial, no card | Free | $0 | — | One-off |
| 50,000 | $1.00, no pack | Starter | $50 | $1.00 | 6 months |
| 500,000 | $1.00, no pack | Standard | $375 | $0.75 | 6 months |
| 2,500,000 | Volume tier | Scale | $1,250 | $0.50 | 6 months |
| 12,500,000 | Talk to sales | Ultimate | $3,750 | $0.30 | 6 months |
Serper pack prices and credit counts as published on serper.dev, October 2026; per-1,000 figures are price divided by credits. Note that these are per credit, not per query — Serper charges 1 credit for up to 10 results and 2 credits for 11–100, so a rank-tracking job that reads past position 10 pays double the rate shown. Credits are valid for six months from purchase, so a pack you only half consume doubles your effective rate: $3,750 spread over 7.5M credits actually used is $0.50 per 1,000, not $0.30. Scrapeless lists $1 per 1,000 successful requests with no pack, no expiry and no deep-result surcharge, dropping below $0.50 per 1,000 on volume plans. Shopping data is billed at twice the base rate. Confirm current rates with both vendors before you commit.
Run the numbers on your own volume. The free trial takes a minute to start and needs no card.
Not price — Serper competes hard there. These are the three places the two products actually diverge.
Serper calls itself the world's fastest Google search API and publishes results in 1–2 seconds. Scrapeless publishes under one second. On a 20,000-keyword daily run that difference is the gap between a job that finishes before the morning standup and one that is still going.
Serper sells prepaid credit packs that expire six months after purchase, so the advertised rate only holds if you consume the whole pack in time. Scrapeless charges for successful requests as you make them — no pack, no expiry, and a failed request is retried at our cost rather than billed to you.
Serper's published endpoints cover Search, Images, News, Maps, Places, Videos, Shopping, Scholar, Patents and Autocomplete — Google AI Overview is not among them. Scrapeless ships a dedicated actor that returns the AI answer as Markdown with every cited source, which is the dataset GEO and brand-visibility work runs on.
Vendor claims are easy to write. These are figures from our own runs — dates, sample sizes and the caveats included. Serper.dev was not included in these particular runs, so what follows is what Scrapeless recorded, plus one head-to-head against Oxylabs.
We timed seven SERP APIs against the same 10 US-English long-sentence queries. Scrapeless returned 10 of 10 usable requests and posted the lowest average, p50 and p95 of the seven. Serper.dev was not one of the seven APIs in that run, so we have no head-to-head timing against it and will not invent one. Serper publishes 1–2 seconds on its own homepage; the figures below are ours, measured. Long-sentence queries are harder than typical short keywords — the published sub-second figure refers to standard queries, and this run is the harder end of the range.
Measured Performance
Average, usable requests
p50, usable requests
p95, usable requests
Usable requests returned
Disclosed in full: Oxylabs returned an internal status 613 on 82 of its 800 requests in this run. That anomaly is part of why the gap is this wide, and we show it rather than leaving it out. This is one configuration on one day — it does not establish that Scrapeless is more reliable across every market, account and setup.
Requests with usable organic results, of 800
Usable-request latency — P50
Usable-request latency — P95
Beyond response speed, we measure the depth and completeness of search results — from organic result coverage and raw HTML retrieval to snippet field availability.
92 / 0
Walking start=0..90 returned 92 organic result items across ten consecutive pages with no empty page. Same run: SerpApi 88, SearchAPI.io 87.
99.5%
In the volume run, 99.5% of returned organic result items carried a snippet field. This is field coverage, not snippet accuracy.
1,818 KB
Fetching metadata.rawUrl as a second request returned roughly 1,818 KB of original HTML, and the document-structure check passed.
Understand how the benchmarks were conducted, what the metrics mean, and the limitations behind the results.
seven SERP APIs against the same 10 US-English long-sentence queries, 10 usable requests each. A ten-query sample is small — treat it as indicative, not definitive.
800 web-search requests per vendor, US English. Figures are for usable requests only.
These are our own tests on our own accounts. Both free tiers are open, so the honest check is your keyword list, your markets, timed on your side.
The 92 figure counts organic result items, not 92 unique destination websites.
Raw HTML needs a second call to metadata.rawUrl. It is not returned inline with the first response.
99.5% is field coverage — the share of result items carrying a snippet — not a measure of whether the snippet text is correct.
A single num=100 request returned HTTP 400 in the same testing; the ten-page walk above is the method that worked.
Seven countries returned organic results, but some responses did not confirm region or language — that is coverage, not geo-targeting accuracy.
Both free tiers are open — run your keyword list through each and time it yourself.
Set the comparison aside for a moment. This is what the Scrapeless Google Search API gives you as a product — the reasons teams stay after the migration, not just the reason they arrive.
A flat, published base rate that drops below $0.50 per 1,000 on volume tiers — with no prepaid pack to buy and no credits that expire before you use them.
Typical response time is under 1 second, which is what makes a daily — or hourly — tracking cadence practical rather than something you batch overnight and read tomorrow.
Billing counts successful requests only. A request that fails is retried at our cost and never reaches your invoice, so your effective cost per usable row is the number you were quoted.
Shape any request by country, language, location and device with gl, hl, google_domain and a proxy country — the same call, pointed at whichever market you need.
Organic results, knowledge graph, local pack and map, latest posts, pagination and search metadata arrive as typed JSON — and AI Overview comes back as Markdown you can hand straight to a model.
Concurrency-based metering with dedicated concurrency on volume plans, a free trial with no credit card, and a named solutions engineer once you are running production workloads.
Where the two services genuinely differ, and where they do the same job. Everything in the Serper.dev column comes from its own public pricing and documentation pages.
| Feature | Scrapeless Google Search API | Serper.dev |
|---|---|---|
| Entry rate per 1,000 | $1.00 per 1,000 successful requests, no pack required | $1.00 per 1,000 credits on the $50 Starter pack |
| Volume rate | Below $0.50 per 1,000 on volume plans — talk to sales | Down to $0.30 per 1,000 credits, on a $3,750 prepaid pack |
| Billing model | Per successful request, pay as you go | Prepaid credit packs |
| Credit expiry | None Nothing to consume before a deadline | 6 months Paid credits lapse six months after purchase |
| Results 11–100 | 1× Standard request, standard rate | 2 credits Deep result sets cost double |
| Failed requests | Not billed — retried at our cost | Not addressed on the pricing page |
| Published response timeStandard queries, vendor stated | Under 1 second | 1–2 seconds |
| Measured response timeour long-sentence benchmark | 2.168s average, p50 1.981s, p95 3.190s — lowest of the seven APIs tested | Not included in that run |
| Free allowance | Yes Free trial, no credit card | Yes 2,500 queries, one-off, no card |
| Google AI Overview | One call Dedicated scraper.overview actor returning Markdown, source[], products[] | Not listed Absent from the published endpoint list |
| Organic results, knowledge graph, local pack | Yes organic_results, knowledge_graph, local_results, local_map, pagination | Yes Structured JSON |
| Published endpoint breadth | Focused Deep on Google search surfaces; other sources via Web Unlocker and Agent Browser | Broad Search, Images, News, Maps, Places, Videos, Shopping, Scholar, Patents, Autocomplete |
| Enterprise tier & SLA | Available on enterprise agreements — ask during onboarding | Not published on the pricing page |
Serper.dev column compiled from its public pricing and documentation pages, October 2026. Vendors update their products; re-check anything that will drive a purchase decision.
Four steps from signing up to a scheduled production job. No SDK is required — the API is one HTTP endpoint, so curl is enough to see your first result.
Sign up with no credit card and you land in the dashboard with a usable key. The free trial is there so you can run your real keyword list before deciding anything. Try for Free →
Find it under API Key in the dashboard. Every request carries it as an x-api-token header — there is no OAuth dance and no key in the query string. API reference →
POST to /api/v1/scraper/request with the actor scraper.google.search and your query in input. A 200 hands back the parsed JSON directly; a 201 returns a taskId to poll for longer jobs. Quickstart →
Vary gl, hl and google_domain to cover each market, page with start and num, and call scraper.overview when you also need the AI Overview answer. Then put it on a cron and watch the cost per run. Product page →
# 1 — organic results for one market
curl 'https://api.scrapeless.com/api/v1/scraper/request' \
--header 'x-api-token: YOUR_KEY' \
--header 'Content-Type: application/json' \
--data '{
"actor": "scraper.google.search",
"input": {
"q": "best running shoes",
"hl": "en",
"gl": "us",
"google_domain": "google.com",
"start": 0,
"num": 20
}
}'
# → organic_results[], knowledge_graph,
# local_results, pagination, search_information
# 2 — the AI Overview answer, same key
curl 'https://api.scrapeless.com/api/v2/scraper/execute' \
--header 'x-api-token: YOUR_KEY' \
--header 'Content-Type: application/json' \
--data '{
"actor": "scraper.overview",
"input": { "prompt": "best running shoes",
"country": "us" }
}'
# → content (Markdown), source[], products[]Full parameter list, response schemas, language examples and the pricing calculator.
Both services take a query and return structured JSON, so the migration is a request rewrite and a field remap. Here is the same search on each side.
What changes in your code: the header name becomes x-api-token, the host becomes api.scrapeless.com/api/v1/scraper/request, and your query parameters move inside an input object under "actor": "scraper.google.search".
What does not change: you are already sending a JSON POST body, and q, hl, gl and num keep their names and meanings — so this is the smallest migration of any SERP vendor pair. The response is a JSON object keyed by result type, so your parser needs a rename rather than a rewrite. A 200 carries the data directly; a 201 returns a taskId to poll.
Get a key, point one job at the new endpoint, and keep the old one running alongside it.
One published rate that falls with volume. No prepaid pack, no seat fees, no per-engine surcharge — and you are billed only for requests that actually return data.
Run your own keyword list through the real API before you decide anything.
The published base rate, charged per successful request rather than a monthly bucket you have to forecast.
Rates drop below half the base price on qualifying volume plans, with capacity and support to match.
Not sure which tier you land in? Start on the free trial, run a week of your real workload, and we will quote from the usage rather than a guess.
· In every plan ·
Requests that return shopping data are billed at twice the base rate. Volume pricing depends on committed monthly volume — see the pricing page for current tiers, or for a quote against your actual query mix.
Pick the workload closest to yours and see the fields, the request shape and the cadence it implies.
Daily positions across keywords, domains and markets.
Local results and map placement per city, at the granularity multi-location brands need.
Whether your brand appears in the AI answer, and which sources it cites.
Shopping surfaces, competitor listings and demand signals from the SERP.
Verify PPC and Ads visibility, placement and competitive messaging across target queries and markets.
The complete scenario library.
Start free without a credit card, run the same queries through each, and compare the response times and the real cost per usable row rather than the headline rate.