Google Search API comparison
FlashData vs Serper.dev
Keep your search tool. Give it more to work with.
Moving from Serper? Compare the cost of the queries you will actually use, preserve your agent’s output, and add YouTube evidence or media delivery to the same workflow.
01 · At a glance
Feature-by-feature comparison
Keep familiar search fields, start with a smaller credit purchase, and add saved jobs or YouTube evidence when your workflow needs them.
| Feature | FlashData | Serper.dev |
|---|---|---|
| Start with a smaller commitment | ||
| Minimum credit purchase | $10$1.00 / 1,000 · pay as you go | $50.00$1.00 / 1,000 · Starter |
| New-account search allowance | Up to 2,000 searchesOne-time signup allowance | 2,500 free queriesSignup offer |
| Agent trial before owner signup | YesUp to 50 Google searches · 72 hours | Sign up and use an account API key |
| Preserve your search tool’s behavior | ||
| Web, images, news, shopping, videos and places | YesSix Google sources | YesCorresponding Google endpoints |
| Country, language and local targeting | gl, hl and locationInside params | gl, hl and locationTop-level request fields |
| Date filters for news and research | Yesparams.tbs · for example qdr:w | Yestbs · for example qdr:w |
| Next page and requested result count | params.page and params.numPage number starts at 1 | page and numPage number starts at 1 |
| Titles, source URLs and snippets | Organic result arraytitle, link, snippet and position | Organic result arraytitle, link, snippet and position |
| Questions, answers and entity context | YesPeople Also Ask, answer box and knowledge graph when returned | YesPeople Also Ask, answer box and knowledge graph when returned |
| Tools for agents and background work | ||
| Official standalone CLI | YesLinux, macOS and Windows · no runtime to install | Not listedIn the public product catalog |
| Markdown search results from the API | YesChoose JSON or Markdown content | JSON search results |
| Persisted async jobs you can resume by ID | YesSubmit, save the ID, retrieve later | Not listedPublished search flow returns realtime results |
| Google queries in one JSON submission | 100 on pay as you goPersisted Batch · up to 500 by plan | Up to 100 queriesMini-batch returns responses together |
| Completion callbacks for background jobs | Yescallback_url on each async or Batch query | Not listedIn the published search flow |
| Owner claim and credential handoff | YesClaim once; reuse the saved CLI profile | Configure an account API key |
| Continue from search into YouTube | ||
| Look up metadata by YouTube video ID | YesDedicated YouTube Metadata API | Not listedGoogle video search is available |
| Timestamped YouTube captions | YesCaption text and timing for video evidence | Not listedIn the public product catalog |
| YouTube video / audio delivery to your storage | YesYouTube Download API | Not listedIn the public product catalog |
The starting purchase uses FlashData pay as you go and Serper Starter. Larger Serper packs have lower unit rates; compare them in the calculator. The agent trial is separate from signup offers and subject to eligibility. Not listed refers to Serper’s public product catalog; community integrations may exist.
02 · Why switch?
What would make switching worthwhile?
Start with the part of your workload that needs to change. You can move Google organic search without replacing every tool in your application.
Your search volume is uneven
A low pack rate helps when you use the queries. Compare the upfront purchase and six-month usage with FlashData’s pay-as-you-go option.
Check pack utilization →Your agent needs evidence from video
A Google video result gets you a URL. FlashData can continue with YouTube metadata, timestamped captions and delivery to your storage.
See the complete tasks →Your search tool is part of a larger system
Keep the agent’s tool input and citation format while moving the provider call. Add persisted jobs when work must resume after a process exits.
Check your integration →03 · Packs & spend
A lower pack rate only helps if you use it.
Choose your volume and a plan on each side. Compare unit price, upfront purchase and included usage, then check how much of a Serper pack you would use before its six-month expiry.
Compare the rate you consume with the amount you purchase. A prepaid pack is not a monthly allowance.
FlashData · Pay As You Go
$1.00per 1,000 Google Search requests
- Monthly fee
- $0.00
- Included requests / month
- 0
- Credit top-up
- From $10.00
- Modeled monthly cost
- $10.00
- Effective cost / 1,000 at this volume
- $1.00
No subscription or included credits. Add prepaid credits as needed.
25 concurrent requests · 100 queries / Batch
Serper.dev · Starter
$1.00per 1,000 queries at this pack’s rate
- Monthly fee
- $0.00
- Queries per prepaid pack
- 50,000
- Pack purchase
- $50.00
- Usage consumed / month
- $10.00
Starting with no credits: $50.00 for 1 pack covers this month, leaving 40,000 queries. Packs are valid for six months.
50 queries / second · no monthly query refill
One pack · six months
Will you use this pack before it expires?
Suppose you buy one Starter pack now and run 10,000 searches every month for six months.
To use the whole pack: about 8,334 searches / month.
- Pack price ÷ queries used, per 1,000
- $1.00
You use the whole first pack within its validity, so its effective cost matches the advertised pack rate. Later packs are outside this forecast.
Concurrency and queries per second measure different capacity limits.
How the estimate works
Select each provider independently. FlashData’s monthly cost is its monthly fee plus search usage above the included credit value. Included requests assume all shared credits go to Google Search. Lowest modeled monthly cost selects the FlashData plan with the smallest total at your volume. Effective monthly cost per 1,000 is that total divided by the entered monthly searches, multiplied by 1,000. Serper’s unit price is the pack purchase divided by its query count, multiplied by 1,000. Usage value shows credits consumed; it is not a monthly subscription charge. Starting with no Serper credits, the purchase estimate rounds up to whole packs at the selected rate; unused queries remain in the purchased packs, which expire after six months. The first-pack forecast assumes one pack bought now and the same monthly volume for six months: queries used = min(pack size, monthly volume × 6); effective cost per 1,000 = pack price ÷ queries used × 1,000. Later pack purchases are outside that forecast. At zero searches no pack purchase is needed and effective cost is undefined. FlashData self-service top-ups start at $10 and may leave a credit balance. Trial credits, existing balances, taxes and other API usage are excluded.
All FlashData Google Search plans
| Plan | Per 1,000 requests | Monthly fee | Included credits | Google Search requests |
|---|---|---|---|---|
| Pay As You Go | $1.00 | $0 · no monthly fee | None | Top up as needed |
| Micro | $0.80 | $49.00 | 1,000 ($50.00 of usage) | ≈ 62,500 |
| Starter | $0.65 | $199.00 | 4,000 ($200.00 of usage) | ≈ 307,692 |
| Pro | $0.60 | $499.00 | 10,000 ($500.00 of usage) | ≈ 833,333 |
| Scale | $0.50 | $999.00 | 20,000 ($1,000.00 of usage) | ≈ 2,000,000 |
| Business | $0.40 | $1,999.00 | 40,000 ($2,000.00 of usage) | ≈ 5,000,000 |
Included credits renew each billing period. The request allowance uses the full shared balance for Google Search.
Full FlashData pricing →All Serper.dev prepaid packs
| Prepaid pack | Per 1,000 queries | Pack purchase | Queries per pack | Monthly fee | Queries / second |
|---|---|---|---|---|---|
| Starter | $1.00 | $50.00 | 50,000 | $0 | 50 |
| Standard | $0.75 | $375.00 | 500,000 | $0 | 100 |
| Scale | $0.50 | $1,250.00 | 2,500,000 | $0 | 200 |
| Ultimate | $0.30 | $3,750.00 | 12,500,000 | $0 | 300 |
Each pack is a one-time purchase with six-month credit validity. There is no monthly refill.
FlashData signup trial
Eligible new accounts can try up to 2,000 Google searches after signup and email verification, using the full shared trial balance for search. No card required.
Signup trial details →Serper.dev signup trial
2,500 free queries, with no card required. This is a trial allowance, not a recurring monthly free plan.
FlashData’s separate agent trial →04 · Your existing tool
Keep the contract your agent already uses.
A Serper integration may return a string, a dictionary or documents with links. Find that boundary before changing the provider call.
| Existing integration | Current output | What to preserve |
|---|---|---|
| LangChain · GoogleSerperAPIWrapper | results() returns a dictionary; run() combines selected snippets into a string. | Check which method your chain calls. Returning an array where the next step expects text changes the tool contract. |
| CrewAI · SerperDevTool | Accepts search_query and returns structured results, including optional blocks. | Keep search_query as the tool input. Map the fields your agent reads; replacing SERPER_API_KEY alone cannot switch providers. |
| Haystack · SerperDevWebSearch | Returns documents and links. Documents contain search snippets. | Preserve both outputs and source URLs. Retain the separate content-fetching component when your pipeline needs full pages. |
Keep the tool name and input schema stable. Adapt FlashData’s response inside the tool so the next step receives the format it already understands.
Build the HTTP tool adapter →Try the normalized search example →
Which search types can move?
| Your existing task | Migration decision | FlashData source or next step |
|---|---|---|
| Google web results | Migrate the organic-search call and normalize its output. | google_search |
| Images, News, Shopping, Videos, Places | Map each task to its dedicated source and parser. | google_images, google_news, google_shopping, google_videos, google_places |
| Scholar, Patents, Autocomplete | Keep these calls with your existing provider; FlashData has no direct equivalent in its current catalog. | No direct equivalent |
| Full text from a linked website | Retain your page-fetching and extraction step after search. | A search snippet is not the full article |
Exact request and response mapping
| Change | Serper.dev | FlashData | Migration note |
|---|---|---|---|
| Search request | POST /search | POST /v1/queries/realtime | Use the FlashData data host and source=google_search. |
| API key | X-API-KEY header | X-API-Key header | Header casing is not the difference. Use a FlashData key with google_search permission. |
| Query | q | query | Rename the request field. |
| Country, language, count | gl, hl, num | params.gl, params.hl, params.num | Nest options under params; keep their values. |
| Location and time filter | location, tbs | params.location, params.tbs | Preserve the targeting used by your existing task. |
| Organic results | organic[] | results[0].organic[] | Read the FlashData response wrapper after status=done. |
| Source fields | title, link, snippet, position | title, link, snippet, position | Keep your application’s citation fields; normalize link to url if needed. |
| Related queries | relatedSearches[] | results[0].relatedSearches[] | Treat this as optional; consume each query field when present. |
| Related questions | peopleAlsoAsk[] | results[0].peopleAlsoAsk[] | A question can arrive without an answer, snippet or URL. Do not assume every object contains evidence. |
| Answer and entity blocks | answerBox, knowledgeGraph | No guaranteed one-to-one mapping | If your tool requires these blocks, keep that path until its required fields have a supported mapping. |
| News, Images or Places | A different endpoint such as /news | A dedicated source such as google_news | Select the source and use its own parameters and result schema. |
Map only the options and result blocks your task uses. An optional related question is not automatically an answered question with a source.
FlashData Google Search schema →05 · Swap the call
Change the request. Keep your source records.
Run a US English search for openai. Both examples print position, title, URL and snippet, so your application can keep its citation format.
1 Set up FlashData access
Sign up or sign in, complete account setup and create a key in API Keys with google_search permission. Check your available credits. Keep the key in your environment or secret store.
An agent can reuse its saved access or start an eligible trial.
2 Run one search
Python 3, standard library only. Save the complete script under the filename shown in your task folder, then run it. It reads the environment variable or prompts privately for a key. Each script performs one search.
Serper.dev
import getpass, json, os
from urllib.request import Request, urlopen
from urllib.error import HTTPError, URLError
key = os.environ.get("SERPER_API_KEY")
if not key:
key = getpass.getpass("Serper key: ")
payload = {"q": "openai", "gl": "us", "hl": "en", "num": 10}
request = Request("https://google.serper.dev/search",
data=json.dumps(payload).encode(),
headers={"X-API-KEY": key, "Content-Type": "application/json"})
try:
with urlopen(request, timeout=120) as response:
body = json.load(response)
except HTTPError as error:
raise SystemExit(f"HTTP {error.code}; check access and request input.") from None
except (URLError, TimeoutError):
raise SystemExit("No response; check state before retrying.") from None
if body.get("error"):
raise SystemExit("Search failed; inspect its error before retrying.")
items = body.get("organic", [])
print(json.dumps([{
"position": item.get("position"), "title": item.get("title"),
"url": item.get("link"), "snippet": item.get("snippet")
} for item in items], ensure_ascii=False, indent=2))Run: python3 serper-search.py
FlashData
import getpass, json, os
from urllib.request import Request, urlopen
from urllib.error import HTTPError, URLError
key = os.environ.get("FLASHDATA_API_KEY")
if not key:
key = getpass.getpass("FlashData key: ")
payload = {"source": "google_search", "query": "openai",
"params": {"gl": "us", "hl": "en", "num": 10}}
request = Request("https://data.flashdata.dev/v1/queries/realtime",
data=json.dumps(payload).encode(),
headers={"X-API-Key": key, "Content-Type": "application/json"})
try:
with urlopen(request, timeout=120) as response:
body = json.load(response)
except HTTPError as error:
raise SystemExit(f"HTTP {error.code}: check input and access.") from None
except (URLError, TimeoutError):
raise SystemExit("No response; check state before retrying.") from None
if body.get("status") != "done":
raise SystemExit("Search is not complete; check its status.")
items = (body.get("results") or [{}])[0].get("organic", [])
print(json.dumps([
{"position": item.get("position"), "title": item.get("title"),
"url": item.get("link"), "snippet": item.get("snippet")}
for item in items
], indent=2, ensure_ascii=False))Run: python3 flashdata-search.py
3 Read the same application fields
Serper’s organic array is at the top level. FlashData’s is inside results[0]; check status === "done" first. Both examples normalize link to url.
An empty array is a valid result. Titles, order and snippets come from the live search.
FlashData Google Search schema →[
{
"position": 1,
"title": "OpenAI | Research & Deployment",
"url": "https://openai.com/",
"snippet": "..."
}
]Before changing your production tool
| Case | What to check |
|---|---|
| Representative searches | Use your existing country, language and location. Check required fields and usable citations; do not compare live rankings byte for byte. |
| Empty or partial results | An empty organic list is valid. Missing optional blocks must not become invented answers or uncited evidence. |
| Access and credit failures | Surface authentication, permission and balance errors. An error must not look like a successful search with no matches. |
| Rate limits or a lost response | Use bounded retries for an explicit rate limit. Inspect saved job state after a timeout; avoid launching a second provider automatically for an uncertain request. |
Start with one workflow. Keep the existing provider configuration available for a deliberate rollback; sending every request to both providers adds another billable search.
Move a recurring or background workload
An async method in your Python wrapper handles waiting inside that process. For work that must survive a restart, submit a persisted FlashData async job, save its ID and retrieve that job later. Batch groups known queries; Bulk processes a file.
06 · Beyond search
What happens after your agent finds a link?
Choose the next task, then the sources it needs. Google Search, YouTube Data and YouTube Download share one account balance, with separate rates and permissions.
Build a research brief with video evidence
Start withA company, product or research question
- Google Search: discover sources
- YouTube Search: find relevant videos
- Metadata + captions: collect timestamped evidence
GetA brief with source links and video timestamps.
Keep web-page extraction for full articles. If a video has no captions, record that gap instead of treating the title as a transcript.
Build the research workflowDeliver a selected video to your storage
Start withA YouTube video selected by your application
- Metadata: confirm the selected video
- Download: submit a delivery job
- Storage: keep the delivered object details
GetVideo or audio in your configured storage, ready for the next job.
Use owner access with youtube_download permission. Temporary agent trial access does not include Download. Delivery uses its own byte-based pricing.
Build the delivery pipelineEnrich a file of known video IDs
Start withAn existing list of YouTube video IDs
- Prepare a text file with one video ID per line
- Bulk: submit once and save the job ID
- Results: retrieve the output and inspect failed items
GetA dataset your application can join back to the original IDs.
Skip Google Search when you already know the IDs. Persist the job ID so a restarted process resumes the existing work.
Process a file with Bulk07 · Agent rollout
Move one tool before moving the whole agent.
Use the migration prompt for a coding agent, or run the checks from an existing CLI profile. Keep your current provider available until the replacement returns the output your workflow needs.
- Inspect the existing tool
Record its search types, filters, required fields and text-or-record output.
- Prove one replacement
Check local fixtures, run one search, and save source records and usage.
- Approve ongoing access
If using a trial, the owner claims it and approves the required permissions and shared balance.
- Resume with saved state
Reuse the owner credential and one run ID per occurrence. Stop when access is revoked.
Current agent trial: 1 credit, up to 50 Google searches, 72 hours. This separate starting allowance is subject to eligibility. Claimed access uses the owner’s account balance.
Give your agent the migration task
Read https://flashdata.dev/compare/serper/reference.md and https://flashdata.dev/auth.md.
Inspect my Serper tool, including its input, country/language filters, output type and consumed fields.
Report unsupported search types or required answer/entity blocks before changing the integration.
Migrate one supported Google organic-search task; retain specialized calls and existing credentials.
Preserve the tool name, input schema, text-or-record output and source URLs.
Reuse existing FlashData access; otherwise check eligibility before starting one trial.
Keep keys in the host secret store, outside chat and source control.
Check normal, empty, missing-optional-field and error responses with local fixtures.
Run one search for "openai" with gl=us, hl=en and num=10; save the result and usage.
Do not automatically call both providers or retry an uncertain submission.
Do not buy credits. For recurring use, ask the owner to claim and approve access.
Reuse the same profile and state directory. Stop on revocation or insufficient credits.Validate with an existing CLI profile
Complete CLI installation and authentication first. These commands use the default profile and submit one search after the free checks.
flashdata auth status --json --non-interactive &&
flashdata search "openai" --gl us --hl en --num 10 \
--dry-run --json --non-interactive &&
flashdata search "openai" --gl us --hl en --num 10 \
--run-id serper-migration-first --json --non-interactiveThe CLI reuses a completed run only when its saved state remains available. Use a new ID for a new occurrence. Your own HTTP tool needs its own persisted state; a CLI run ID is not an API idempotency header.
Stop on revoked access or insufficient credits. Claiming does not buy credits or create a separate spending cap.
Codex →OpenClaw →Trial and claiming →Recurring tasks and revocation →
08 · Questions before switching
The details worth settling first.
I already bought a Serper pack. Do I have to move everything?
No. Move one supported task first and retain other calls while you use your existing pack. Serper credits do not transfer to FlashData. Keep separate provider credentials and route each task deliberately.
Which price should I compare: the pack rate or the cost of queries I use?
Use both. The pack rate assumes you consume the whole pack; the utilization forecast shows what happens if you do not. For a FlashData subscription, compare the monthly fee and included usage at your actual volume, not just its lower unit rate.
Will changing SERPER_API_KEY switch my existing agent tool?
No. A Serper-specific tool still calls Serper’s endpoint and parses its response. Replace the provider call, use a FlashData key, and adapt the response to the same tool output. An MCP connection is a separate integration choice, not a replacement value for that environment variable.
Will every result and rich search block be identical?
Live results can differ in order and content. Migrate the fields your application actually consumes. Organic title, URL, snippet and position have a clear mapping; optional answers and entity blocks need their own mapping before you switch that path.
Does an async Python method give me a resumable job?
An async client method lets your process do other work while waiting for a response. A persisted FlashData async job has a job ID you can save and retrieve later. Use the latter when work must survive a process restart, and keep that ID before scheduling another submission.
Does one balance mean the same price for search, captions and downloads?
No. Products share credits but have separate rates, billing units and permissions. A Google search is billed as a search request; media delivery uses byte-based pricing. Approve the required owner permissions before adding a new product to a recurring workflow.
When is keeping Serper the simpler choice?
When your existing tool already does the whole job, you use your prepaid pack within its validity, or you rely on Scholar, Patents, Autocomplete or provider-specific fields without a FlashData equivalent. You can keep those calls and use FlashData for the additional workflows you need.
Start with one search
Connect FlashData to your next task.
Run the example, or give your agent the migration prompt.