Answers
How do I compare data API prices quoted per call, per result, per credit or per GB?
Short answer
Convert every quote to the cost of 1,000 usable items. For per-call pricing that is the price times the calls needed: okdata Xiaohongshu search is $0.02 for about 20 notes, so $1.00 per 1,000. For per-result pricing it is the price times 1,000, plus any per-run fees. Credits, compute units and proxy gigabytes need one more step, so test a real run before trusting the estimate.
Vendor A says $0.02 a request, vendor B says $1 per 1,000 results, vendor C sells credits, and a scraping platform bills compute units plus proxy traffic. None of those numbers can be compared directly. The one number that can: what 1,000 items you can actually use will cost.
The formulas
| Quoted as | Cost per 1,000 items | Ask the vendor |
|---|---|---|
| Per call (per page) | price × ⌈1,000 ÷ items per page⌉ | How many items does a page really hold? Are empty pages billed? |
| Per result | price × 1,000, plus per-run or start fees | What counts as a result: replies, duplicates, ads? |
| Credits | credits per call × price per credit × calls | Do credits expire? Does the credit price change by plan? |
| Compute units | compute per run × unit price, plus proxy and storage | How much does a typical run of my size use? |
| Proxy per GB | GB per 1,000 pages × price per GB, plus your servers and time | How heavy are the pages? How many requests fail and retry? |
Scraping platforms such as Apify bill compute units, proxies and storage for some scrapers and a fixed price per result for others, so the same platform can land in two rows. Residential proxy networks sell traffic by the gigabyte, and a page with video previews weighs far more than a text page.
okdata in this table
Most endpoints are per call, so cost per 1,000 is printed next to each one: Xiaohongshu note search is $1.00, TikTok video comments $0.04. One endpoint is per item: Official Account search charges $0.058 per account returned, at least one and at most $0.58 per call. Every response states its own charge:
curl -s -G "https://api.okdata.online/v1/wechat_mp/accounts/search" \
-H "Authorization: Bearer $OKDATA_API_KEY" \
--data-urlencode 'query=36氪' | jq '.meta.billing, (.data.items | length)'
The traps that wreck comparisons
- Usable items. If a third of what a vendor returns are duplicates from overlapping pages, its real cost per 1,000 is 50% higher.
- Failures. Ask whether failed requests, timeouts and blocked pages are billed. With proxies you always pay for the traffic, success or not.
- Fields. A cheap result without like counts or full text may force a second, pricier lookup.
- Your time. Compute and proxy pricing leaves the fixing to you when a platform changes.
Run each shortlisted option on the same 1,000-item job and divide the real invoice by the usable items. The pricing page lists cost per 1,000 for every endpoint here.
Updated