The useful SERP API cost is the spend required to produce an acceptable result, not just the advertised price of a search call. Count searches, selected-page reads, charged add-ons, retries that incur a charge, and unused prepaid balance. Keep the denominator explicit: searches requested, successful searches and usable workflow results can produce different numbers.
For the procurement decision across billing horizons, use payment-model and unused-quota calculation rather than compare an advertised unit rate alone.
Calculate the complete request mix
For a SearchCans workflow, count one credit for a charged single-page SERP operation and two for a charged standard Reader extraction. A batch page=N request fetches pages one through N and adds one credit for each additional page; count charged page units rather than just HTTP calls. Selecting a specific p page is a different request mode. Add the selected Reader proxy credits: shared adds two, datacenter adds five, and residential adds ten. A rendering mode does not itself add another credit charge. Use current documentation to distinguish charged operations from uncharged failed attempts.
credits = charged_serp_page_units + 2 * reader_calls + proxy_addon_credits
workflow_cost = credits * acquisition_cost_per_credit
cost_per_usable_result = workflow_cost / usable_resultsAn illustrative mixed-workflow calculation
Suppose a local worksheet contains ten charged searches and twenty charged Reader operations. Four reads use the shared proxy add-on and two use the datacenter add-on; the rest use no proxy. This is an illustrative workload, not an observed production benchmark or a provider recommendation.
| Operation | Credits |
|---|---|
| Ten standard searches | 10 |
| Twenty standard Reader operations | 40 |
| Four shared proxy add-ons | 8 |
| Two datacenter add-ons | 10 |
| Total | 68 |
If only five outputs satisfy the application requirement, divide the total workflow cost by five, not by ten or thirty. Obtain acquisition cost from the purchased plan and account for unused or expired balance when reporting effective cost.
Optimize the denominator and the request count
Shortening a payload or converting text to Markdown does not by itself reduce a per-operation credit charge. Token costs belong in a separate model that uses the actual model tokenizer and retained evidence. No universal savings percentage is assumed.
- Deduplicate the same query and market before scheduling it.
- Set cache freshness from the task; reusing outdated evidence is not a usable result.
- Read selected URLs instead of every discovered link.
- Keep failure and retry records so a cheaper total is not hiding missing output.
- Compare a before-and-after workload using the same acceptance rules.
Reference documentation: www.searchcans.com — /pricing; www.searchcans.com — /reader-api.
