Expiring credits, paid empty searches, a free tier that runs out once. Here's the comparison, with receipts.
| CrustAPI | Serper | |
|---|---|---|
| Paid credits expire | Never | After 6 months |
| Free tier | 3,000 / month, no card | 2,500 once, no card |
| Empty results | Free | 1 credit, even for zero results |
| Images per credit | ~100 | 10 default; 100 costs 2 credits |
| Maps data | Counts, hours, sites, stars | Lean place fields |
| Entry price | $49 for 25,000 | $50 for 50,000 |
| Batch (100 queries) | In the dashboard | API + dashboard |
| MCP for AI agents | Official (crustapi-mcp) | None official |
| CLI | crust | None official |
| Agents can self-pay | x402, no signup | No |
| Reverse image (Lens) | Not yet | Yes |
Serper figures from their public site and our paid probes of their API, July 11, 2026. Anything stale? Email support@crustapi.com and we'll fix it.
On July 11 we sent the same nonsense query to both APIs. A search that finds nothing should cost nothing, and one of them agrees.
Raw timings, both billing probes, and the exact script: the benchmark file. Re-run it yourself.
Same result arrays, same field names, same extras like knowledgeGraph and peopleAlsoAsk. We built it that way on purpose, so your parser, schema, and prompts keep working.
GET instead of POST, type= instead of a URL per endpoint, lowercase header. That's the migration.
Serper credits expire six months after purchase; their public Trustpilot reviews include a customer describing about $1,000 expiring. Ours never expire, and our terms say so in writing.
One credit per successful search; zero results and failures cost nothing. If 12 of your 100 lead-gen queries come back empty, you pay for 88.
Serper's 2,500 free queries are one-time. Ours refill: 3,000 free credits every month, no card, so a small project may never pay at all.
Maps includes review counts, hours, websites, and an opt-in star distribution serper doesn't sell. Shopping adds a condition field, and images are drawn below.
We measured 99 on July 11. Routinely 85 to 100.
Default. num=100 returns 100 images for 2 credits, so 50 per credit at best.
Identical queries, same machine, medians. Serper won two of the four rows, and here they are anyway.
Both live in the one-second band, so switch for the economics, not speed. Raw data and the script: benchmarks/serper-crustapi-2026-07-11.txt.
Per raw credit at entry, serper is cheaper: $50 buys 50,000 there, $49 buys 25,000 here. But say you buy in January, use 20,000, and the project pauses:
$50 paid, 20,000 usable searches. The rest burned on the six-month clock.
$49 paid, every credit eventually used. Leftovers wait for the next project.
Image workloads flip harder: covering 100 images costs 1 credit here, 2 there at best. Packs go self-serve all the way up, no sales calls: 25k $49 · 100k $149 · 500k $549 · 2.5M $1,999 · 10M $6,500 · up to 250M $100,000, ex-tax.
When serper is the better pick: you run a steady, high-volume, web-search-only firehose and reliably burn every credit inside six months (their raw rate at scale is hard to beat), or you need reverse image search today. We don't serve Lens yet.
We're one option, not the only one. If neither of us fits:
Point your real queries at both APIs for a day. It costs nothing on our side, and the results decide, not this page.
Get 3,000 free credits