At list price, Serpex is the cheapest web search API per 1,000 searches for an AI agent in this comparison, at $0.65 on a $100 top-up. Serper is next at $1.00 per 1,000, and Parallel and Perplexity also charge $1.00 in their fast modes. A Google search on cloro's Hobby plan is $1.20. Per correct answer the order changes: in an open benchmark of six web search APIs, cloro had the lowest cost, $1.98 per 1,000 correct answers, against $5.11 to $15.08 for the other five.
- List price: ten APIs cost from $0.65 to $15.00 per 1,000 searches. Four cost less than $2. Brave, Parallel and Perplexity cost $5 in their default modes, Exa $7, Tavily $8 and SerpApi $15 on its $75 plan.
- The rate is not the bill: more results, a deeper mode or an AI summary can double the price of one search. A search that does not hold the answer costs a second search.
- Cost per correct answer is known for six APIs only. The benchmark did not test Serpex, Serper, Perplexity or Firecrawl.
What does each web search API cost per 1,000 searches?
The basis is the live list price on each vendor's own pricing page, read on November 14, 2026. Where the vendor sells pay as you go, the table uses it. Otherwise it uses the plan or pack closest to $100 a month. Each row is one real-time search in the default mode, with 10 results. No row uses a batch or queued rate.
| API | What you buy | $ per 1,000 searches | Cheaper mode | Results from |
|---|---|---|---|---|
| Serpex | Pay as you go, $100 top-up ($50 buys 76,923 credits); 1 credit a search | $0.65 | No | Several sources, not named |
| Serper | Pack: $50 for 50,000 credits; 1 credit a search | $1.00 | No | |
| cloro | Hobby: $100 a month for 250,000 credits; a Google search is 3 credits | $1.20, or $2.00 with the AI Overview (5 credits) | No | |
| Firecrawl | Standard: $99 a month for 100,000 credits; 2 credits per 10 results | $1.98 | No | Its own search endpoint |
| Brave | Pay as you go | $5.00 | No | Brave's own index |
| Parallel | Pay as you go, advanced mode (the default) |
$5.00 |
fast or turbo, $1.00 |
Parallel's own index |
| Perplexity Search API | Pay as you go | $5.00 |
fast, $1.00 |
Perplexity's index |
| Exa | Pay as you go, auto type |
$7.00 |
lite or instant, $4.00 |
Exa's neural index |
| Tavily | Pay as you go, $0.008 a credit; a basic search is 1 credit | $8.00 | No | Tavily's index |
| SerpApi | Developer: $75 a month for 5,000 searches | $15.00 | No | Google and other engines |
The Hobby row in numbers: $100 for 250,000 credits is $0.40 per 1,000 credits. A Google search is 3 credits, so 1,000 searches use 3,000 credits and cost $1.20. With the AI Overview a search is 5 credits, so 1,000 searches cost $2.00. The smaller Lite plan, $30 for 37,500 credits, comes to $2.40 per 1,000 Google searches.
Most of these rates go down with volume. The table keeps one rate per vendor so that the rows compare like for like.
What changes the bill besides the list price?
The rate per search is the start of the bill. Five things move it:
- Results per search. Firecrawl charges 2 credits per 10 results. Exa and Parallel charge $1 per 1,000 results after the first 10.
- Search depth. A Tavily advanced search is 2 credits, so $16 per 1,000 on pay as you go. Exa's deep types cost $12 to $15 per 1,000.
- The AI summary. On a Google SERP API the AI Overview can cost extra. On the Hobby row above it takes the rate from $1.20 to $2.00.
- Extra billed searches. SerpApi bills a second search when part of the Google page loads separately. In the benchmark below, that happened on 97% of short factual questions, so one question cost about 1.97 searches.
- Failed and expired credits. Perplexity does not bill invalid, rate-limited or failed requests, but it bills a successful response with no results. Serpex does not bill failed requests. Serper's packs expire after 6 months.
Which web search API is cheapest per correct answer?
An agent needs correct answers, so the cost that counts is the cost per correct answer. Six web search APIs ran on the same questions in November 2026: one search per question, 10 results, one reader model and one grader. Cost per correct answer is the price per 1,000 searches divided by the accuracy on 1,000 short factual questions (SimpleQA Verified).
| API | $ per 1,000 searches (benchmark basis) | $ per 1,000 correct answers | Multi-hop accuracy (FRAMES) |
|---|---|---|---|
| cloro | $1.92 | $1.98 | 80.2% |
| Parallel | $5.00 | $5.11 | 65.3% |
| Brave | $5.00 | $5.37 | 53.9% |
| Tavily | $5.00 | $5.45 | 48.4% |
| Exa | $7.00 | $7.19 | 64.0% |
| SerpApi | $7.25 | $15.08 | 77.4% |
These prices use a different basis from the first table. The benchmark priced each API on the plan that covers 100,000 searches a month, read on November 5, 2026. Its Google row is an async search with the AI Overview (5 credits) on the Starter plan, $250 a month for 650,000 credits. Tavily is $5 there because its Growth plan is $500 a month for 100,000 credits; pay as you go, it is $8. SerpApi is $7.25 on its Searcher plan, $725 a month for 100,000 searches. On November 14, 2026, these plan rates had not changed.
Two limits apply to this table. On short factual questions, accuracy is close: Parallel (97.8%), Exa (97.3%) and the top Google API (96.9%) are within the margin of error. If your agent looks up single facts, list price and speed decide, and Exa was the fastest at 0.74 seconds median against about 8 seconds for the two Google APIs. Also, the benchmark did not test Serpex, Serper, Perplexity or Firecrawl. A $1.00 search that holds the answer as often as a $1.92 search is cheaper per correct answer. The benchmark harness is one Python file, so you can run it on your own questions with your own keys.
Which one should an AI agent use?
| Your agent needs | Cheapest fit at list price |
|---|---|
| The lowest price per call, from any source | Serpex or Serper |
| Answers to multi-hop questions from one search | A Google SERP API with the AI Overview, the lowest cost per correct answer in the benchmark |
| Sub-second searches | Parallel fast or turbo, or Perplexity fast, at $1.00 per 1,000 |
| Page text in the response, for RAG | Firecrawl, Exa or Tavily |
| An index that is not Google | Brave, Exa or Parallel |
For a few hundred searches a month, the free tiers cost nothing: Tavily gives 1,000 credits, Exa $10 of credit and Brave $5 of credit a month. Past a few thousand searches a month, the paid rate and the share of searches that hold the answer decide the bill.
How this was checked
This comparison is published by cloro, which is one of the APIs in it. Every other price comes from that vendor's own pricing page or pricing docs, read on November 14, 2026, and each one is linked in the first table. Its own row comes from its published plans. The accuracy and the cost per correct answer are copied from the benchmark linked above, which publishes its questions, settings and every graded answer. The full comparison, with free tiers and what each API returns, is in Best Web Search API for AI Agents.
Top comments (0)