Web Search API Pricing for Agents: $0.65 to $15 per 1,000
Ten web search APIs for AI agents run $0.65 to $15 per 1,000 searches, but cost per correct answer reorders the table. Full pricing breakdown.

Ten web search APIs for AI agents run from $0.65 to $15.00 per 1,000 searches at list price, and four of them cost less than $2.
A comparison whose list prices were read on November 14, 2026 puts Serpex cheapest at $0.65 per 1,000 on a $100 top-up, Serper next at $1.00, with Parallel and Perplexity also at $1.00 in their fast modes. A Google search on cloro's Hobby plan is $1.20. One caveat up front: the comparison is published by cloro, which is one of the APIs in it.
The list price
Every price was read from the vendor's own pricing page on November 14, 2026, using pay-as-you-go where offered and otherwise the plan closest to $100 a month — one real-time search, default mode, 10 results.
| API | $ / 1,000 searches | Cheaper mode |
|---|---|---|
| Serpex | $0.65 | No |
| Serper | $1.00 | No |
| cloro (Hobby) | $1.20 | No |
| Firecrawl | $1.98 | No |
| Brave | $5.00 | No |
| Parallel | $5.00 | fast/turbo, $1.00 |
| Perplexity | $5.00 | fast, $1.00 |
| Exa | $7.00 | lite/instant, $4.00 |
| Tavily | $8.00 | No |
| SerpApi | $15.00 | No |
The rate is not the bill
Four things move the number. Depth: a Tavily advanced search is 2 credits, so $16 per 1,000 pay-as-you-go, and Exa's deep types run $12 to $15 per 1,000. Summaries: an AI Overview takes the cloro Hobby row from $1.20 to $2.00.
Extra billed searches: SerpApi bills a second search when part of the Google page loads separately, which the benchmark saw on 97% of short factual questions — about 1.97 searches per question. And failure policy: Perplexity does not bill invalid, rate-limited or failed requests but does bill a successful response with no results, Serpex does not bill failed requests, and Serper's packs expire after 6 months.
Cost per correct answer reorders the table
An agent needs answers, not calls. In an open benchmark of six APIs on 1,000 short factual questions, cost per 1,000 correct answers came out at $1.98 for cloro, $5.11 Parallel, $5.37 Brave, $5.45 Tavily, $7.19 Exa and $15.08 SerpApi. Multi-hop accuracy on FRAMES ranged from 80.2% down to 48.4%.
Two limits on that table. On short factual questions accuracy is close enough to be a tie — Parallel 97.8%, Exa 97.3%, the top Google API 96.9%, inside the margin of error — so for single-fact lookups, price and speed decide, and Exa was fastest at 0.74 seconds median against about 8 seconds for the two Google APIs. The benchmark also did not test Serpex, Serper, Perplexity or Firecrawl, which includes the two cheapest rows above.
What to pick
If your agent does single-fact lookups at volume, the cheap list-price rows are doing the same job as the expensive ones — start at Serpex or Serper and measure. If it chains multi-hop questions, pay for the accuracy, because a search that misses the answer costs you a second search anyway. If you need page text back for RAG, that narrows you to Firecrawl, Exa or Tavily regardless of price.
Under a few hundred searches a month, free tiers cover it: Tavily gives 1,000 credits, Exa $10 of credit and Brave $5 of credit a month. Past a few thousand, run the benchmark harness — it is one Python file — on your own questions before committing to a plan.
More from DangMua