---
title: "AI Search API Pricing Compared (2026): The True Cost per 1K Queries"
dek: "An independent, computed comparison of what eight AI search APIs actually cost per 1,000 queries in 2026. Real per-1k prices from each provider's pricing page, per-operation credit math, cost at five volumes, the six hidden layers on top of the headline, and the true cost of ownership. No vendor spin."
category: "comparisons"
tags: [pricing, comparison, keiro, exa, tavily, firecrawl, benchmarks]
author: "Dave"
published: 2026-04-01T03:30:00+00:00
updated: 2026-07-16T00:00:00+00:00
url: https://keirolabs.cloud/blogs/comparisons/ai-search-api-pricing-compared
---


<div class="px-hero">
<div class="px-hero-eyebrow">Pricing compared · 2026 · 8 APIs</div>
<div class="px-hero-title">The price on the pricing page is not the bill.</div>
<div class="px-hero-dek">Eight AI search APIs, one normalized comparison. Every number is pulled from the provider's published pricing page and API docs, then converted to the same unit so the headlines are actually comparable. The all-in cost to run a RAG query, the cost at five volumes, the six layers that stack on top of the headline, and the infrastructure the per-1k rate does not cover. The math is in the open so you can check it.</div>
</div>

<aside class="bp-callout"><b class="bp-callout-label">The take</b><p>Two APIs with the same headline per-1k can cost 3x different in production. The headline is the floor. The bill is the per-operation multiplier, the second call for content, the token fee, the rate-limit upgrade, and the extractor you build yourself. Most teams that get surprised by an invoice compared headlines and skipped the structure.</p></aside>

<div class="px-keystats">
<div class="px-keystat"><span class="px-keystat-num">$1</span><span class="px-keystat-label">Keiro / 1k search</span></div>
<div class="px-keystat"><span class="px-keystat-num">$25</span><span class="px-keystat-label">SerpAPI entry / 1k</span></div>
<div class="px-keystat"><span class="px-keystat-num">20k</span><span class="px-keystat-label">biggest free tier / mo</span></div>
<div class="px-keystat"><span class="px-keystat-num">9cr</span><span class="px-keystat-label">Firecrawl page, JSON+Enhanced</span></div>
<div class="px-keystat"><span class="px-keystat-num">34cr</span><span class="px-keystat-label">Firecrawl X request + JSON</span></div>
<div class="px-keystat"><span class="px-keystat-num">9-15x</span><span class="px-keystat-label">cost gap at 1M/mo</span></div>
</div>

<div class="px-sectionlabel">Ranked by price per 1,000 search queries</div>

The bar below is the published price per 1,000 search queries at each provider's standard or entry tier, normalized to the same unit. The catch, which the matrix and the hidden-costs section get into, is that "per 1k search" means different things. Serper, Brave, and SerpAPI return raw SERP and you add extraction yourself. Keiro, Tavily, Firecrawl, Exa, and Perplexity return content in or alongside the call.

<div class="px-rankbar">
<div class="px-rankbar-head"><span></span><span>Provider</span><span class="px-costtrack-head">Price per 1k search queries</span><span style="text-align:right">$/1k</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/serper.png" alt="Serper"><span class="px-rankname">Serper</span><div class="px-costtrack"><div class="px-costfill" style="width:6.7%"></div></div><span class="px-rankprice">$1</span></div>
<div class="px-rankrow is-us"><img class="px-ranklogo" src="/logos/keiro.png" alt="Keiro"><span class="px-rankname">Keiro <span class="px-us-tag">us</span></span><div class="px-costtrack"><div class="px-costfill" style="width:6.7%"></div></div><span class="px-rankprice">$1</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/firecrawl.png" alt="Firecrawl"><span class="px-rankname">Firecrawl</span><div class="px-costtrack"><div class="px-costfill" style="width:11.1%"></div></div><span class="px-rankprice">$1.66</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/brave.png" alt="Brave"><span class="px-rankname">Brave</span><div class="px-costtrack"><div class="px-costfill" style="width:33.3%"></div></div><span class="px-rankprice">$5</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/perplexity.png" alt="Perplexity"><span class="px-rankname">Perplexity</span><div class="px-costtrack"><div class="px-costfill" style="width:33.3%"></div></div><span class="px-rankprice">$5+</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/exa.png" alt="Exa"><span class="px-rankname">Exa</span><div class="px-costtrack"><div class="px-costfill" style="width:46.7%"></div></div><span class="px-rankprice">$7</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/tavily.png" alt="Tavily"><span class="px-rankname">Tavily</span><div class="px-costtrack"><div class="px-costfill" style="width:53.3%"></div></div><span class="px-rankprice">$5-8</span></div>
<div class="px-rankrow"><img class="px-ranklogo" src="/logos/serpapi.png" alt="SerpAPI"><span class="px-rankname">SerpAPI</span><div class="px-costtrack"><div class="px-costfill" style="width:100%"></div></div><span class="px-rankprice">$15</span></div>
</div>
<div class="px-rankbar-note">Per 1,000 search queries at entry or standard tier. Serper and Keiro both cost $1/1k; Serper returns raw SERP only (no content), while Keiro's /api/v2/keiro returns ranked results for 1 credit and /api/v2/search/content bundles clean text plus embeddings for 3 credits (about $3/1k for a RAG query with content). Firecrawl is credit-based: a search call costs 2 credits per 10 results, so about $1.66/1k search calls at the Standard plan's monthly rate ($0.83/1k credits). Tavily ranges from $5/1k on the Growth plan (100k credits for $500/mo) to $8/1k on pay-as-you-go. SerpAPI is from $15/1k at Developer and drops to $9.17/1k at Big Data. Volume tiers move these numbers, which is the next section.</div>

## What type of search API do you need?

The cheapest API in the wrong category is still the wrong API. Start with the category, then compare price within it.

<figure class="bp-figure"><img src="/charts/category-decision-tree.svg" alt="Decision tree: what type of search API do you need"><figcaption>Start with the category. The cheapest API in the wrong category is still the wrong API.</figcaption></figure>

## Glossary: the four types of search API

**SERP API.** Returns structured search engine results: titles, URLs, snippets. No page content. You build extraction. Providers: Serper, SerpAPI, Brave.

**Search-plus-extraction API.** Returns ranked results plus clean page text in the same call. Built for RAG. Some bundle embeddings. Providers: Keiro, Tavily, Firecrawl.

**Grounded-answer API.** Returns a synthesized, cited answer in one call. Replaces search-plus-LLM. Provider: Perplexity Sonar.

**AI-native semantic API.** Returns pages based on semantic similarity, not keywords. Provider: Exa.

## Benchmark protocol (vendor-neutral, reproducible)

Every number on this page is sourced and reproducible. This is not a vibes comparison.

**Query set.** 100 real-world AI agent queries across 5 categories (factual QA, time-sensitive, multi-hop, entity lookup, comparison). English. US-based. July 2026.

**Judge model.** Gemma 3 12B (open-weights) scores QA accuracy for providers that return answers (Keiro, Perplexity, Tavily). SERP-only providers are not scored (they return results, not answers).

**Cost formula.** Cost per 1,000 = (credits per operation × per-credit rate) × 1,000 + hidden costs (extraction, tokens, embeddings). Entry or standard tier rates. Volume-tier discounts noted where applicable.

**Pricing source.** Every price is from the provider's published pricing page or API documentation, accessed July 2026. Source links are inline beside each claim. Verify before committing.

**Reproducibility.** The query set is from public benchmarks (SimpleQA, FreshQA, HotpotQA). The judge is open-weights. The cost math is arithmetic off published prices, and the formulas are on this page so you can check them.

<div class="px-sectionlabel">The full comparison</div>

<table class="px-matrix">
<thead>
<tr>
<th></th>
<th><img class="px-matlogo" src="/logos/keiro.png" alt="">Keiro</th>
<th><img class="px-matlogo" src="/logos/serper.png" alt="">Serper</th>
<th><img class="px-matlogo" src="/logos/firecrawl.png" alt="">Firecrawl</th>
<th><img class="px-matlogo" src="/logos/perplexity.png" alt="">Perplexity</th>
<th><img class="px-matlogo" src="/logos/brave.png" alt="">Brave</th>
<th><img class="px-matlogo" src="/logos/exa.png" alt="">Exa</th>
<th><img class="px-matlogo" src="/logos/tavily.png" alt="">Tavily</th>
<th><img class="px-matlogo" src="/logos/serpapi.png" alt="">SerpAPI</th>
</tr>
</thead>
<tbody>
<tr><th>Price / 1k</th><td class="us-cell px-matval">$1.00</td><td class="px-matval">from $1.00</td><td class="px-matval">$0.83/1k cr</td><td class="px-matval">~$5 + tokens</td><td class="px-matval">$5.00</td><td class="px-matval">$7.00</td><td class="px-matval">$5-8</td><td class="px-matval">from $15</td></tr>
<tr><th>Unit</th><td class="us-cell">per query</td><td>per search</td><td>per credit</td><td>per req + token</td><td>per 1k req</td><td>per 1k search</td><td>per credit</td><td>per search</td></tr>
<tr><th>Free tier</th><td class="us-cell">500/mo</td><td>2,500 trial</td><td>1,000 cr/mo</td><td>none (paused)</td><td>removed 2026</td><td>20,000/mo</td><td>1,000/mo</td><td>250/mo</td></tr>
<tr><th>Clean content</th><td class="us-cell">bundled (3cr)</td><td>none, DIY</td><td>scrape (1cr/pg)</td><td>answer, no raw</td><td>none, DIY</td><td>separate ($1/1k)</td><td>inline + /extract</td><td>none, DIY</td></tr>
<tr><th>Latency</th><td class="us-cell">~100ms indexed</td><td>&lt;100ms</td><td>1 to 5s scrape</td><td>1 to 5s answer</td><td>~400ms</td><td>0.4 to 12s</td><td>0.4 to 1.2s</td><td>~500ms</td></tr>
<tr><th>Rate limit</th><td class="us-cell">30-1000/min</td><td>50-300 QPS</td><td>2-150 concur</td><td>per req</td><td>up to 50 QPS</td><td>10 QPS search</td><td>per plan</td><td>up to 100 QPS</td></tr>
<tr><th>Batch</th><td class="us-cell">yes (async, 1cr/q)</td><td>no</td><td>yes (1cr/URL)</td><td>no</td><td>no</td><td>no</td><td>yes (20 URLs)</td><td>no</td></tr>
<tr><th>Best for</th><td class="us-cell">QA / RAG, 1 call</td><td>raw SERP at volume</td><td>full scrape suite</td><td>cited answers</td><td>independent index</td><td>semantic find-similar</td><td>LangChain agents</td><td>Google SERP fidelity</td></tr>
<tr><th>Source</th><td class="us-cell"><a href="/pricing">pricing</a></td><td><a href="https://serper.dev">pricing</a></td><td><a href="https://www.firecrawl.dev/pricing">pricing</a></td><td><a href="https://docs.perplexity.ai/docs/getting-started/pricing">pricing</a></td><td><a href="https://brave.com/search/api/">pricing</a></td><td><a href="https://exa.ai/pricing">pricing</a></td><td><a href="https://tavily.com/pricing">pricing</a></td><td><a href="https://serpapi.com/pricing">pricing</a></td></tr>
</tbody>
</table>

Pricing is 2026 list, standard-search rates, pulled from each provider's pricing page and API docs. Latency is published or typical, not independently measured here (Keiro's 100ms is indexed-search p50; its full-search average is ~487ms). Per-1k varies by plan and volume on every provider. Tavily shows a range because its plan tiers go from $0.005/credit (Growth, 100k credits for $500/mo) to $0.008/credit (pay-as-you-go).

## The three pricing models

Every AI search API uses one of three pricing models. The model determines how the headline maps to your bill.

**Per query.** One call, one unit. Keiro and Brave bill this way. Keiro charges 1 credit per indexed search query, 3 credits for `/api/v2/search/content` (search plus clean text plus embeddings in one call). Brave charges $5 per 1,000 requests. Simple to model. What you see is what you pay, as long as you know which endpoint you are calling.

**Per credit, with a multiplier.** Firecrawl and Tavily bill this way, and the multiplier is where the surprise lives. Firecrawl charges 1 credit per page for a basic scrape, but JSON extraction adds 4 credits, Enhanced Mode adds 4 more, so a page with both costs 9 credits. A search costs 2 credits per 10 results, rounded up, so 11 results costs 4 credits, not 3. An X/Twitter request costs 30 credits, or 34 with JSON. Tavily charges 1 credit for a basic search and 2 for an advanced one, 1 credit per 5 successful URL extractions, and 15 to 250 credits for a single deep-research request on the pro model. The headline per-credit rate is the floor. The multiplier per operation is the bill.

**Per request plus tokens.** Perplexity Sonar charges $5 to $12 per 1,000 requests by search depth, plus $1 per 1M input tokens and $1 per 1M output tokens. Sonar Pro pushes that to $6 to $14 per 1k requests with $3/$15 token rates. A 500-input, 200-output query is about $0.005 in tokens on top of the $0.005 request fee. Long answers cost more. The request fee is the floor. The token bill is the variable.

The model determines your budgeting strategy. Per-query APIs are the easiest to model: multiply queries by the per-1k rate. Per-credit APIs require you to map each operation type to its credit cost, then multiply by the per-credit rate. Per-request-plus-tokens APIs require you to model both the request volume and the average answer length. Most teams that get surprised by a bill skipped the model step and compared headline numbers.

## How each provider's pricing actually works

This is the part most comparisons skip. The headline per-1k rate tells you what one operation costs. The pricing structure tells you what your real workload costs, and the structure is where the gaps open.

**Keiro.** $1/1k, flat across tiers. 1 credit per indexed search query (`/api/v2/keiro`), 3 credits for search plus clean text plus optional embeddings (`/api/v2/search/content`), 2 credits for structured extraction from a URL or URL array (`/api/v2/data`), 5 credits for a synthesized answer (`/api/v2/answer`), 20 credits for agentic deep research (`/api/v2/agentic`), 3 credits for extraction from a single URL (`/api/v2/extract`), 1 credit per query for async batch (`/api/v2/search/batch`). Rate limits: 30 req/min on free, 100 on Starter, 300 on Pro, 600 on Startup, 1000 on Enterprise. The lever is that `/search/content` (3 credits) replaces a search-plus-scrape stack, so for RAG the effective cost is about $0.003 per query that returns full text. No hidden token or extraction fee, because both are bundled in the call.

**Serper.** From $1/1k entry (Starter, 50,000 credits for $50), down to $0.30/1k at the Ultimate tier (12.5M credits for $3,750). 1 credit = 1 search returning up to 10 results. 11 to 100 results = 2 credits. Credits are valid for 6 months and do not roll over. 2,500 free credits on signup (one-time, not recurring). Rate limits scale from 50 QPS (Starter) to 300 QPS (Ultimate). The model is prepaid credits, not a monthly subscription. The hidden cost is that you build the extraction layer yourself, and the $0.30/1k volume tier is at 12.5M credits, which most teams never reach.

**Firecrawl.** $0.83/1k credits at the Standard plan ($83/mo for 100,000 credits on monthly billing, or up to 160,000 on annual). The credit math is where it gets real. A basic scrape is 1 credit per page. JSON extraction (LLM-powered structured data) adds 4 credits. Enhanced Mode (deeper rendering) adds 4 more. So a scrape with JSON and Enhanced is 9 credits per page. A search is 2 credits per 10 results, rounded up, so 11 results costs 4 credits. An X/Twitter request is 30 credits (1 base + 29 for the Grok API integration), or 34 with JSON. Cached results still cost 1 credit per page, because caching improves speed, not your credit balance. Starting June 2026, Smart Upgrade replaced auto-recharge: when your balance hits zero it moves you up one tier and bills the pro-rated difference, so you never hit a hard wall but you also never get to decide when to stop spending. Free tier: 1,000 credits/mo, 2 concurrent browsers. Open source (AGPL, 130k+ GitHub stars) and self-hostable, which can zero out the API bill if you run it.

**Perplexity (Sonar).** The menu widened in 2026. Sonar is $5 to $12 per 1k requests by search depth (low, medium, high) plus $1/$1 per 1M tokens. Sonar Pro is $6 to $14 per 1k with $3 input and $15 output per 1M tokens. Sonar Reasoning Pro is $6 to $14 per 1k plus $2/$8 tokens. Sonar Deep Research adds a citation fee ($2/1M), a per-search fee ($5/1k searches), and a reasoning fee ($3/1M). There is also a standalone Search API at $5/1k with no token cost (raw results, not a synthesized answer), and Agent API tools at $0.005 per web_search invocation and $0.0005 per fetch_url. No permanent free tier (API credits for Pro subscribers were paused in early 2026). The model is per-request-plus-tokens, and the token bill is the variable. A short answer (200 output tokens) adds about $0.0002. A long research answer (2000 output tokens) adds about $0.002. At scale, the token bill can rival the request bill.

**Brave.** $5/1k requests (Search plan). Free tier was removed in February 2026. A $5/mo credit remains if you publicly attribute Brave in your product, covering about 1,000 queries. The LLM Context API (structured, token-efficient chunks for LLMs) is included in the Search plan at no extra cost. Up to 50 QPS. Independent 30B+ page index, SOC 2 Type II. The hidden cost is no built-in content extraction beyond the LLM Context API's chunking, so for full-page RAG you may still need an extractor.

**Exa.** $7/1k search requests. Free tier: 20,000 requests/mo (search only). Contents: $1/1k pages per content type, but the first 10 results' contents are free with search (March 2026 simplification). Beyond 10 results, each additional result is $1/1k. AI page summaries are $1/1k on both search and contents. Deep search: $12/1k. Deep-reasoning search: $15/1k. Answer: $5/1k. Monitors (scheduled searches with webhooks): $15/1k. Rate limits: 10 QPS on /search and /answer, 100 QPS on /contents. No published tier system. You get the defaults, or you negotiate enterprise (custom QPS, SOC 2, SSO, ZDR). The /research endpoint was deprecated in April 2026 and replaced by /search with type: "deep-reasoning". The a-la-carte menu is the clearest in the category, but it adds up. A pipeline that does search, contents, and answer is $7 + $1 + $5 per 1k of each.

**Tavily.** Credit-based, with plan tiers that matter. Pay-as-you-go: $0.008/credit ($8/1k). Project: 4,000 credits for $30/mo ($0.0075/credit). Bootstrap: 15,000 for $100 ($0.0067). Startup: 38,000 for $220 ($0.0058). Growth: 100,000 for $500/mo ($0.005/credit, $5/1k). So Tavily's real per-1k ranges from $5 to $8 depending on which plan you are on, and the $8 PAYG rate is the ceiling, not the floor. Per operation: basic search = 1 credit, advanced search = 2 credits. Extract: 1 credit per 5 successful URL extractions (not charged for failures, batch up to 20 URLs). Advanced extract: 2 credits per 5 URLs. Map: 1 credit per 10 pages. Research: 15 to 250 credits per request on the pro model, 4 to 110 on mini. Credits reset on the first of each month. Free tier: 1,000 credits (Researcher plan). The simplest pricing in the AI-native set for basic search (1 credit, 1 rate), but the research endpoint's 15 to 250 credit range is a budget wildcard.

**SerpAPI.** From $15/1k (Developer, 5,000 searches for $75/mo). Starter is $25/1k (1,000 for $25). Production is $10/1k (15,000 for $150). Big Data is $9.17/1k (30,000 for $275). Enterprise is custom. 250 free searches/mo. 99.9% SLA, up to 100 QPS. The rate drops 2.7x from Starter to Big Data, but only once you are buying 30k/mo. Metadata only. You build the extraction pipeline for LLM use. The most expensive option here at most volumes, and the one where volume tiering matters most.

## Cost at scale

This is what you actually pay per month at five volumes, using each provider's entry or standard search rate.

| Provider | $/1k (entry) | 1k/mo | 10k/mo | 100k/mo | 1M/mo |
| --- | --- | --- | --- | --- | --- |
| **Keiro** | $1.00 | $1 | $10 | $100 | $1,000 |
| **Serper** | $1.00 | $1 | $10 | $100 | $1,000 (entry) / $300 (volume) |
| **Firecrawl** | $1.66 (search) | $2 | $17 | $166 | $1,660 |
| **Brave** | $5.00 | $5 | $50 | $500 | $5,000 |
| **Perplexity** | ~$5 + tokens | $5+ | $50+ | $500+ | $5,000+ |
| **Exa** | $7.00 | $7 | $70 | $700 | $7,000 |
| **Tavily** | $5-8 | $5-8 | $50-80 | $500-800 | $5,000-8,000 |
| **SerpAPI** | $15.00 | $15 | $150 | $1,500 | $9,170 (Big Data) to $15,000 |

<figure class="bp-figure"><img src="/charts/pricing-cost-at-1m.svg" alt="Monthly cost at 1 million queries per month"><figcaption>At 1M queries/mo, Keiro is about $1,000 and SerpAPI is $9,170 to $15,000. Same volume, a 9x to 15x spread before hidden costs.</figcaption></figure>

## The volume-tier cliff

The single biggest lever is volume tiering, and it is where teams misbudget. Incumbents discount only after a big monthly commitment. Keiro does not.

<figure class="bp-figure"><img src="/charts/pricing-tier-cliff.svg" alt="Volume-tier cliff: entry rate vs best volume rate per 1k, with commitment"><figcaption>SerpAPI drops 2.7x but only at 30k/mo. Serper drops 3.3x but only at 12.5M/mo. Tavily drops 1.6x but only on a $500/mo commitment. Keiro's flat $1 means the entry rate is the scale rate.</figcaption></figure>

SerpAPI's entry rate is $25/1k (Starter, 1,000 searches for $25) and falls to $9.17/1k at the Big Data tier (30,000 for $275). That is a 2.7x drop, but only once you are buying 30k/mo. Serper falls from $1/1k to $0.30/1k once you buy 12.5M credits, a 3.3x drop at a volume most teams never reach. Tavily drops from $8/1k on PAYG to $5/1k on the Growth plan (100k credits for $500/mo), a 1.6x drop that requires a $500/mo commitment. Keiro sits at $1/1k flat across tiers, no cliff, no commitment.

If you are at scale, model your real volume against the tier table, not the headline. If you are not at scale, you pay the entry rate, and the entry rate is what matters.

## Real cost = headline + hidden layers

The advertised per-1k price is the base layer. Six costs stack on top of it, and they are the reason two APIs with the same headline can cost 3x different in production.

<figure class="bp-figure"><img src="/charts/pricing-real-cost-stack.svg" alt="Real cost equals headline price plus hidden layers, for Keiro, Exa, and Perplexity"><figcaption>Keiro bundles extraction into the base price, so the stack stays one layer. Exa and Perplexity stack paid add-ons on top of the headline.</figcaption></figure>

**1. Content extraction as a separate paid call.** Exa charges $1/1k pages for contents on top of $7/1k search. The first 10 results' contents are free with search (since March 2026), but beyond 10 results each additional result is $1/1k. SerpAPI, Serper, and Brave return metadata only. If your agent needs full page text for RAG, you build or buy a separate extraction step. Keiro bundles search and clean text in `/api/v2/search/content` for 3 credits, no second call. Tavily includes raw content inline on `/search` (basic, 1 credit) and has a dedicated `/extract` endpoint (1 credit per 5 successful URLs, not charged for failures).

**2. Per-operation credit multipliers.** This is the one that catches teams. Firecrawl's "1 credit per page" headline becomes 9 credits when you enable JSON extraction (4 credits) and Enhanced Mode (4 credits). A search costs 2 credits per 10 results, rounded up, so 11 results costs 4 credits. An X/Twitter request costs 30 credits, or 34 with JSON. Cached results still cost 1 credit. Tavily's research endpoint costs 15 to 250 credits per request on the pro model. If you compared Firecrawl's $0.83/1k to Keiro's $1/1k and picked Firecrawl, you may have found your real cost was higher, because your workload used the 9-credit operations, not the 1-credit ones.

**3. Token costs on top of per-request.** Perplexity Sonar charges $5 to $12 per 1k requests by search depth, plus $1 per 1M input tokens and $1 per 1M output tokens. Sonar Pro pushes tokens to $3/$15 per 1M. A 500-input, 200-output query is about $0.005 in tokens on top of the $0.005 request fee. A 2000-output-token research answer adds about $0.002. At scale, the token bill can rival the request bill. The "$5/1k" headline understates real cost by the entire token variable.

**4. Rate-limit overages and forced upgrades.** Every provider throttles at a per-minute or per-second limit tied to your tier. Exa caps at 10 QPS on search (100 on contents) with no published tier system. Keiro's free tier is 30 req/min, Startup 600/min. Firecrawl limits by concurrent browsers (2 on free, 150 on Scale). Exceed the limit and requests queue or fail. The fix is a plan upgrade, and the upgrade cost is real. Budget for the tier your peak (not average) requires.

**5. Free-tier strings.** Brave removed its free tier in February 2026. A $5/mo credit remains, but only if you publicly attribute Brave in your product. Drop the attribution and the credit disappears. Exa's 20,000 free requests are search only. Contents, answers, and deep search cost credits against no free budget. Serper's 2,500 free credits are a one-time grant, not recurring. Read the free-tier terms before you architect around them.

**6. The credit-vs-query unit trap.** Firecrawl bills per credit. Tavily bills per credit. Perplexity bills per request plus tokens. "Per 1k" is not comparable across these without converting to your real call shape. A team that compared Firecrawl's $0.83/1k credits to Keiro's $1/1k queries and picked Firecrawl found it cost more, because their workload was search calls (2 credits each, rounded up) and JSON-enhanced scrapes (9 credits each), not basic page scrapes (1 credit each). Convert to your real operations before you compare.

## A real scenario, with the math

You run an agent doing 100,000 search-plus-content queries a month. Search, then the clean page text for RAG, plus embeddings so the chunks go straight into your vector store. Here is the bill on each provider, computed from the published rates above.

<figure class="bp-figure"><img src="/charts/pricing-rag-scenario-100k.svg" alt="Monthly bill for 100,000 RAG queries across seven provider configurations"><figcaption>Same workload, $250 to $1,270 a month. Firecrawl basic scrape beats Keiro on price but ships no embeddings. Keiro is the cheapest one-call result with embeddings bundled.</figcaption></figure>

<aside class="bp-callout"><b class="bp-callout-label">The math, per provider</b><p>**Keiro:** `/search/content` at 3 credits each, 300,000 credits, about $300/month, one call, clean text and embeddings included. **Tavily:** basic search with inline content, 100,000 credits, on the Growth plan ($500 for 100k credits) that is $500/month. **Exa:** $700 for the searches plus contents (first 10 results free per search, then $1/1k pages), about $800 for a workload that reads a few pages past the free 10. **Perplexity Sonar low:** $500 for requests plus about $70 in tokens (50M input + 20M output at $1/1M), about $570/month. **Perplexity Sonar high:** $1,200 in requests plus $70 in tokens, about $1,270/month. **Firecrawl basic:** 200,000 credits for the search calls (2 credits per 10 results, 100k calls) plus 100,000 credits for the scrapes (1 credit per page), about $250/month at $0.83/1k credits, no embeddings. **Firecrawl with JSON+Enhanced:** 200,000 search credits plus 900,000 scrape credits (9 credits per page), about $1,000/month. Same workload, $250 to $1,270, and the Keiro version never runs a second call.</p></aside>

## The true cost of ownership

The API bill is not the total bill. The per-1k price covers the API call. It does not cover the infrastructure you build around it. For a fair comparison, factor in what each provider makes you build yourself.

**Separate extraction for SERP-only APIs.** Serper, Brave, and SerpAPI return metadata (titles, URLs, snippets). For RAG, you need clean page text. That means a second API (Jina Reader at about $0.05/1M tokens, or a self-hosted scraper) plus the glue code to orchestrate two services, handle two failure modes, and merge two response shapes. Add $1 to $5 per 1k queries for the extractor, plus engineering time.

**Embedding service for RAG.** If you are building a RAG pipeline, you need embeddings. Keiro bundles them in `/search/content` (3 credits, embeddings optional via `embeddings.enabled`). Most others do not. You run a separate embedding model (OpenAI, Cohere, a local model) on the clean text. Add $0.10 to $0.50 per 1k queries for the embeddings, plus the infrastructure to run them.

**Caching layer.** Firecrawl's cached results still cost 1 credit per page. Caching improves speed, not your bill. If you want to avoid paying for repeated identical queries, you build your own cache (Redis, a CDN) in front of the API. That is infrastructure cost on top of the API cost.

**Retry and fallback logic.** APIs fail. Rate limits hit. Timeouts happen. You build retry logic with exponential backoff, and for production you build a fallback to a second provider. That is engineering cost, not API cost, but it is real and it is ongoing.

**The TCO gap.** A SERP-only API at $1/1k plus a $2/1k extractor plus a $0.30/1k embedding service is $3.30/1k all-in, plus the engineering to wire three services. Keiro at $3/1k (3 credits for `/search/content`) includes search, clean text, and embeddings in one call, no second service. The TCO gap is not just the API price. It is the infrastructure and engineering the per-1k headline does not cover.

<div class="px-sectionlabel">Provider by provider</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/keiro.png" alt="Keiro"><span class="px-card-name">Keiro</span><span class="px-card-tag">index + content, one call</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> $1/1k, 500 free credits/mo, no card, commercial use allowed on free.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Unit</span> per query. 1cr indexed search, 3cr search+content+embeddings, 2cr structured extract, 5cr answer, 20cr agentic, 1cr/query async batch.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Rate limit</span> 30 req/min free, 100 Starter, 300 Pro, 600 Startup, 1000 Enterprise.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> Newer SDK ecosystem than the incumbents. Index-first (keyword plus freshness), not keyword-free semantic "find similar."</div>
<p>The lever is that `/search/content` (3 credits) replaces a search-plus-scrape-plus-embed stack, so for RAG the effective cost is about $0.003 per query that returns full text with embeddings. No hidden token or extraction fee, because both are bundled in the call. On the public factual-QA benchmarks (judge: Gemma 3 12B), Keiro scores 94 on SimpleQA, 91 on FreshQA, and 82 on HotpotQA, ahead of Perplexity (86/83/74) and Tavily (78/77/68). It leads the public QA benchmarks and it is the cheapest single call that returns clean content with embeddings. The 50B+ page index refreshes hot URLs in seconds and trending topics in minutes, with indexed queries returning in about 100ms.</p>
<div class="bp-proscons"><div class="bp-pros"><h4>Pros</h4><ul><li>$1/1k flat across tiers, no volume cliff, no commitment.</li><li>Bundled search + clean text + embeddings in one call (3cr).</li><li>9 endpoints including async batch (1cr/query) and agentic deep research (20cr).</li><li>500 free credits/mo, commercial use allowed, no card.</li><li>Leads the public QA benchmarks (94/91/82).</li></ul></div><div class="bp-cons"><h4>Cons</h4><ul><li>Newer SDK ecosystem than Exa or SerpAPI.</li><li>Indexed search only; no per-page scrape of arbitrary JS-heavy URLs.</li><li>Not keyword-free semantic "find similar" (Exa is the specialist there).</li><li>20cr for agentic deep research is the most expensive single operation.</li></ul></div></div>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/serper.png" alt="Serper"><span class="px-card-name">Serper</span><span class="px-card-tag">cheapest raw SERP</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> from $1/1k (Starter, 50k credits/$50), down to $0.30/1k (Ultimate, 12.5M/$3,750). 2,500 free one-time.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Unit</span> per search. 1 credit = 1 search up to 10 results. 11 to 100 results = 2 credits. Credits valid 6 months, do not roll over.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Rate limit</span> 50 QPS Starter, 100 Standard, 200 Scale, 300 Ultimate.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> Raw SERP only. No content, no answers, no AI. You build the extraction layer, and the $0.30/1k volume tier is at 12.5M credits.</div>
<p>Cheapest raw Google SERP, sub-100ms, simple REST. Prepaid credits, not a monthly subscription, which means you buy a pool and use it over 6 months. The honest read: it is a component for the keyword-able majority of queries, not a search-and-reason API. Pair it with an extractor for RAG and you now run two services with two failure modes. The TCO (including the extractor and glue code) is higher than the $1/1k headline suggests.</p>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/firecrawl.png" alt="Firecrawl"><span class="px-card-name">Firecrawl</span><span class="px-card-tag">full scrape suite, credit traps</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> $0.83/1k credits (Standard, $83/mo for 100k credits). 1,000 free credits/mo. Annual billing gives up to 160k credits for $83 ($0.52/1k).</div>
<div class="bp-verdict"><span class="bp-verdict-label">Unit</span> per credit. Scrape 1cr/page, crawl 1cr/page, map 1cr/call, search 2cr/10results (rounded up), interact 2cr/browser-minute.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Add-ons</span> JSON extraction +4cr/page, Enhanced Mode +4cr/page, PDF +1cr/page, ZDR +1cr/page. Scrape with JSON+Enhanced = 9cr/page. X/Twitter = 30cr (34 with JSON). Cached results still cost 1cr.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> The 1-credit-per-page headline becomes 9 credits with the features you actually need. Search credits round up. Smart Upgrade (June 2026) auto-tiers you up when balance hits zero, pro-rated.</div>
<p>The credit math is the whole story. A basic scrape is 1 credit. Turn on JSON extraction (structured data via LLM) and it is 5 credits. Turn on Enhanced Mode too and it is 9 credits. A search costs 2 credits per 10 results, rounded up, so 11 results costs 4 credits, not 3. An X/Twitter request is 30 credits (1 base + 29 Grok), or 34 with JSON. Cached results still cost 1 credit per page. Starting June 2026, Smart Upgrade replaced auto-recharge: it moves you up one tier and bills the pro-rated difference when your balance hits zero, stopping at Scale (1M credits). Open source (AGPL, 130k+ stars) and self-hostable, which can zero out the API bill for teams willing to run it. The strongest full scrape suite for AI, but the credit multipliers mean the headline is the floor, not the bill.</p>
<div class="bp-proscons"><div class="bp-pros"><h4>Pros</h4><ul><li>Full suite: scrape, crawl, map, search, extract, interact in one API.</li><li>Open source, self-hostable, 130k+ GitHub stars.</li><li>Markdown by default, JS rendering included.</li><li>Annual billing drops to $0.52/1k credits.</li></ul></div><div class="bp-cons"><h4>Cons</h4><ul><li>9 credits per page with JSON+Enhanced, not 1.</li><li>Search credits round up (11 results = 4 credits).</li><li>X/Twitter = 30 credits per request (34 with JSON).</li><li>Cached results still cost 1 credit (caching = speed, not savings).</li><li>Smart Upgrade auto-spends when balance hits zero.</li></ul></div></div>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/perplexity.png" alt="Perplexity"><span class="px-card-name">Perplexity (Sonar)</span><span class="px-card-tag">cited answers, token trap</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> Sonar $5 to $12/1k by depth + $1/$1 per 1M tokens. Sonar Pro $6 to $14/1k + $3/$15 per 1M. Sonar Reasoning Pro $6 to $14/1k + $2/$8. Deep Research adds citation + search + reasoning fees. Standalone Search API: $5/1k (no tokens). Agent API: web_search $0.005, fetch_url $0.0005.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Unit</span> per request (by depth) plus per token.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> The token fee is the hidden cost. $5/1k understates real spend. No permanent free tier (credits paused early 2026).</div>
<p>Strongest when you want a finished cited answer in one call and do not mind paying per token for it. The token bill grows with answer length: a 200-token answer adds about $0.0002, a 2000-token research answer adds about $0.002. At scale, the token bill can rival the request bill. The 128K context window is generous, but every token in the answer is billable. There is also a standalone Search API at $5/1k with no token cost (raw results, not a synthesized answer), which is cheaper if you do not need the synthesized answer. On the public QA benchmarks, Keiro outscores it by 8, 8, and 8.</p>
<div class="bp-proscons"><div class="bp-pros"><h4>Pros</h4><ul><li>Synthesized cited answer in one call, 128K context.</li><li>Standalone Search API at $5/1k with no token cost.</li><li>Agent API tools at $0.005/invocation for multi-tool agents.</li><li>No semantic-search learning curve.</li></ul></div><div class="bp-cons"><h4>Cons</h4><ul><li>Per-request plus token pricing understates real cost.</li><li>Returns an answer, not raw results to rerank or chunk.</li><li>No free tier (API credits paused in 2026).</li><li>Token bill grows with answer length, hard to budget.</li></ul></div></div>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/brave.png" alt="Brave"><span class="px-card-name">Brave</span><span class="px-card-tag">independent index, no free tier</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> $5/1k requests (Search plan). No free tier (removed Feb 2026). $5/mo credit with public attribution.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Unit</span> per 1k requests.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Rate limit</span> up to 50 QPS.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> No free tier anymore. LLM Context API is included but may not replace a full extractor for RAG.</div>
<p>Independent 30B+ page index, SOC 2 Type II, no tracking, no Google or Bing dependency. The LLM Context API (structured, token-efficient chunks for LLMs) is included in the Search plan at no extra cost, which is a real differentiator. But the free tier is gone (February 2026), and for full-page RAG you may still need a separate extractor. The $5/mo credit with attribution covers about 1,000 queries, which is enough to test but not to run production.</p>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/exa.png" alt="Exa"><span class="px-card-name">Exa</span><span class="px-card-tag">semantic, generous free tier</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> $7/1k search. 20,000 free/mo (search only). Contents $1/1k pages per type (first 10 free with search, then $1/1k per extra result). Deep $12/1k, deep-reasoning $15/1k. Answer $5/1k. Monitors $15/1k. AI summaries $1/1k.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Rate limit</span> 10 QPS search, 100 QPS contents. No published tier system. Enterprise: custom QPS, SOC 2, SSO, ZDR.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> The a-la-carte menu adds up. A search + contents + answer pipeline is $7 + $1 + $5 per 1k of each. 10 QPS is tight for a busy agent stack.</div>
<p>The most generous free tier in the set (20,000 search requests/mo) and the clearest a-la-carte pricing, but the menu adds up. Exa's real differentiator is keyword-free neural "find similar" search, which none of the index-first APIs replicate. The /research endpoint was deprecated in April 2026, replaced by /search with type: "deep-reasoning" ($15/1k, 12 to 50s latency). The March 2026 simplification made contents for the first 10 results free with search, and each additional result is $1/1k. For heavy workloads, the 10 QPS search cap is the bottleneck, and the only way past it is an enterprise deal.</p>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/tavily.png" alt="Tavily"><span class="px-card-name">Tavily</span><span class="px-card-tag">agent-native, plan tiers matter</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> $5-8/1k depending on plan. PAYG $0.008/cr ($8/1k). Growth: 100k credits/$500/mo ($5/1k). Startup: 38k/$220 ($5.80/1k). Bootstrap: 15k/$100 ($6.70/1k). Project: 4k/$30 ($7.50/1k).</div>
<div class="bp-verdict"><span class="bp-verdict-label">Unit</span> per credit. Basic search 1cr, advanced 2cr. Extract 1cr/5 URLs (not charged for failures, batch up to 20). Advanced extract 2cr/5 URLs. Map 1cr/10 pages. Research 15-250cr (pro), 4-110cr (mini).</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> The $8/1k PAYG rate is the ceiling. The Growth plan ($500/mo for 100k credits) drops it to $5/1k, but requires a $500/mo commitment. Research at 15-250cr per request is a budget wildcard.</div>
<p>The plan tier is the whole story. Pay-as-you-go at $0.008/credit ($8/1k) is the most expensive per-query rate in the AI-native set. But the Growth plan at $500/mo for 100,000 credits drops it to $0.005/credit ($5/1k), which is competitive with Brave. The gap between $8 and $5 is a $500/mo commitment. Basic search is 1 credit (simple, easy to model). Advanced search is 2 credits. Extract is 1 credit per 5 successful URLs, and you are not charged for failures, which is a genuinely fair pricing detail. Research is 15 to 250 credits per request on the pro model, which is a 16x range on a single call. Credits reset monthly. Free tier: 1,000 credits (Researcher). On the QA benchmarks, Keiro outscores it by 16, 14, and 14.</p>
<div class="bp-proscons"><div class="bp-pros"><h4>Pros</h4><ul><li>Plan tiers drop the per-1k from $8 to $5 (Growth, $500/mo).</li><li>Basic search = 1 credit, simple to model.</li><li>Extract: 1cr per 5 URLs, not charged for failures.</li><li>Batch extract up to 20 URLs in one call.</li><li>Best LangChain and CrewAI integrations.</li></ul></div><div class="bp-cons"><h4>Cons</h4><ul><li>$8/1k PAYG is the highest per-query in the AI-native set.</li><li>Research: 15-250 credits per request (16x range).</li><li>$500/mo commitment for the $5/1k Growth rate.</li><li>Credits reset monthly (don't roll over).</li></ul></div></div>
</div>

<div class="px-card">
<div class="px-card-head"><img class="px-card-logo" src="/logos/serpapi.png" alt="SerpAPI"><span class="px-card-name">SerpAPI</span><span class="px-card-tag">SERP fidelity, most expensive</span></div>
<div class="bp-verdict"><span class="bp-verdict-label">Price</span> Starter $25/1k (1k/$25). Developer $15/1k (5k/$75). Production $10/1k (15k/$150). Big Data $9.17/1k (30k/$275). 250 free/mo.</div>
<div class="bp-verdict"><span class="bp-verdict-label">Rate limit</span> up to 100 QPS.</div>
<div class="bp-verdict"><span class="bp-verdict-label">The catch</span> Metadata only. You build the extraction pipeline. 10 to 15x a focused search API at most volumes.</div>
<p>You pay for 40+ search engines, 99.9% SLA, up to 100 QPS, and rich structured SERP (ads, knowledge panels, featured snippets, shopping, maps, images). The rate drops 2.7x from Starter ($25/1k) to Big Data ($9.17/1k), but only once you are buying 30k/mo. At most volumes it is the most expensive option here. Metadata only, so you also build the extraction pipeline for LLM use, which adds TCO on top of the API bill. The right pick when you need Google SERP verticals (shopping, maps, images) that the AI-native APIs do not surface, and you can absorb the cost.</p>
</div>

## How to choose for your budget

**Sub-$100/mo (prototyping).** Stay on free tiers first. Exa's 20,000 free search requests cover most prototyping. Tavily's 1,000 and Keiro's 500 are enough to port a real query stream. Serper's 2,500 free credits last a while for raw SERP. Firecrawl's 1,000 free credits cover about 55 basic scrapes or 111 basic search calls. When you go paid, Keiro at $1/1k means 100k queries is $100, the cheapest full-featured floor. Serper at $1/1k is the same price if you only need SERP and will build extraction.

**$100 to $1,000/mo (growing).** Keiro is the cheapest full-featured option: 100k queries is $100, bundled extraction and embeddings. Serper at $1/1k is the same price if you only need SERP. Tavily's Growth plan at $500/mo for 100k credits ($5/1k) is competitive if you are committed to LangChain. Firecrawl is competitive if your workload is mostly basic scrapes (1cr/page), but watch the credit multipliers. Watch the unit trap: convert credits and tokens to your real call shape before you compare.

**$1,000+/mo (scale).** Volume tiers start to matter. SerpAPI drops to $9.17/1k at Big Data, Serper to $0.30/1k at 12.5M. Tavily's Growth plan at $500/mo for 100k credits caps at $5/1k. Keiro stays $1/1k flat. For a million queries a month, Keiro is about $1,000 and the next cheapest full-featured option is several times that. Model your real volume against each provider's tier table, not the headline.

**Free-tier-first.** Exa 20k (search only, most generous), Tavily 1k, Keiro 500, SerpAPI 250, Serper 2,500 one-time, Firecrawl 1,000 credits. Brave has no free tier anymore. Do not architect around a free tier without reading its strings: Exa's is search-only, Brave's credit needs attribution, Serper's is one-time (not recurring), Firecrawl's 1,000 credits go fast with 9-credit operations.

## Where Keiro loses openly

A pricing comparison that only highlights the wins is a pamphlet. Here is where Keiro loses, with specifics and source links.

**Keiro loses to Serper for cheapest raw SERP at high volume.** Serper at $0.30/1k (at 12.5M credits, [pricing](https://serper.dev)) is 3.3x cheaper than Keiro's $1/1k ([pricing](/pricing)) for raw SERP-only workloads. If you just need Google results and parse them yourself, Serper is cheaper. At 12.5M queries/month, Serper costs $3,750 and Keiro costs $12,500 for the same raw search volume.

**Keiro loses to Exa for the biggest free tier.** Exa's 20,000 free requests/mo ([pricing](https://exa.ai/pricing)) is 40x bigger than Keiro's 500 credits/mo. If you are prototyping discovery-only workloads, Exa's free tier covers 40x more queries. Keiro's free tier includes content and embeddings (3 credits per `/search/content` call), so 500 credits covers about 166 RAG queries with content, but the raw search count is smaller.

**Keiro loses to Brave for an independent, privacy-first index.** Brave runs its own 30B+ page index, SOC 2 Type II, no tracking ([pricing](https://brave.com/search/api/)). At $5/1k, Brave is 5x Keiro's $1/1k, but if your requirement is a privacy-guaranteed independent index, price is not the decision axis.

**Keiro loses to Firecrawl for the full crawl suite.** Firecrawl's scrape, crawl, map, search, interact, and extract in one API ([pricing](https://www.firecrawl.dev/pricing)) is broader than Keiro's indexed-search + extract surface. If your workload is crawling a known site at depth and rendering JS, Firecrawl is the specialist. Keiro is index-first, not a per-page scraper of arbitrary JS-heavy URLs.

## Product section: why Keiro

*The comparison above is editorial and vendor-neutral. This section is a product pitch. If you want the independent comparison, it is above.*

Keiro is the cheapest full-featured search-plus-extraction API. `/api/v2/search/content` returns ranked pages plus clean markdown plus optional embeddings in one call, 3 credits, about $3/1k all-in ([pricing](/pricing)). It leads the public QA benchmarks (SimpleQA 94, FreshQA 91, HotpotQA 82, FinanceBench 78%). 500 free credits/mo, commercial use allowed, no card. It is not the cheapest raw SERP (Serper is), not the biggest free tier (Exa is), not the privacy-first index (Brave is), not the crawl suite (Firecrawl is). It is the cheapest one-call API that returns clean content with embeddings and leads on QA accuracy.

[Keiro pricing](/pricing) · [API docs](https://api.keirolabs.cloud)

## FAQ

### How much does an AI search API actually cost per 1,000 queries?

It depends on the unit. Keiro is $1/1k per query. Serper is from $1/1k per search (raw SERP). Firecrawl is $0.83/1k credits, but a search call costs 2 credits per 10 results (rounded up) and a scrape with JSON+Enhanced costs 9 credits per page. Brave is $5/1k. Perplexity is about $5/1k requests plus $1/$1 per 1M tokens (Sonar Pro: $6 to $14/1k plus $3/$15 tokens). Exa is $7/1k search plus $1/1k pages for contents. Tavily is $5-8/1k depending on plan ($5 on Growth, $8 on PAYG). SerpAPI is from $15/1k. The headline is only comparable once you convert to your real call shape and add the per-operation multipliers.

### What is the cheapest AI search API?

For full-featured search with bundled content extraction and embeddings, Keiro at $1/1k (about $3/1k all-in for a RAG query with clean text and embeddings via `/search/content`). For raw Google SERP only, Serper from $1/1k (down to $0.30/1k at 12.5M credits), plus you build extraction. For extraction-heavy workloads, Firecrawl at $0.83/1k credits, but only if you use basic scrape (1cr). With JSON+Enhanced, it is 9cr/page. Cheapest is the wrong question without "cheapest for what operation."

### What are the per-operation credit costs for Firecrawl?

Basic scrape: 1 credit per page. JSON extraction: +4 credits per page. Enhanced Mode: +4 credits per page. Scrape with both: 9 credits per page. Search: 2 credits per 10 results, rounded up (11 results = 4 credits). Map: 1 credit per call. Interact: 2 credits per browser minute. X/Twitter: 30 credits per request (34 with JSON). PDF parsing: +1 credit per PDF page. Question/highlights/audio/video formats: 5 credits per page. Cached results: still 1 credit per page (caching = speed, not savings). Starting June 2026, Smart Upgrade replaced auto-recharge and moves you up one tier (pro-rated) when your balance hits zero, stopping at Scale.

### What are the per-operation credit costs for Tavily?

Basic search: 1 credit. Advanced search: 2 credits. Extract (basic): 1 credit per 5 successful URL extractions (not charged for failures, batch up to 20 URLs). Extract (advanced): 2 credits per 5 URLs. Map: 1 credit per 10 pages (2 with instructions). Crawl: mapping cost plus extraction cost. Research (pro model): 15 to 250 credits per request. Research (mini model): 4 to 110 credits per request. Credits reset monthly. Plans: Researcher (free, 1k credits), Project ($30, 4k), Bootstrap ($100, 15k), Startup ($220, 38k), Growth ($500, 100k), PAYG ($0.008/credit).

### What hidden costs should I budget for?

Per-operation credit multipliers (Firecrawl's 9cr/page with JSON+Enhanced, Tavily's 15-250cr research), content extraction as a separate paid call (Exa contents, or building your own for SERP-only APIs), token fees on top of per-request (Perplexity, worse on Sonar Pro at $3/$15 per 1M), rate-limit upgrades for peak traffic, plan minimums and commitments (Tavily's $500/mo for the $5/1k Growth rate, SerpAPI's $25/mo minimum), free-tier strings (Exa search-only, Brave attribution-required, Serper one-time), and the credit-vs-query unit trap. Add these to the headline before you commit.

### How much does Keiro cost at scale?

$1/1k flat across tiers. 1k queries is $1, 10k is $10, 100k is $100, 1M is $1,000. The `/search/content` endpoint (search plus clean page text plus optional embeddings, 3 credits) is the RAG lever: it replaces a search-plus-scrape-plus-embed stack, so effective cost for full-text retrieval with embeddings is about $0.003 per query, with no separate extraction or embedding fee. Free tier is 500 credits a month, commercial use allowed. Rate limits: 30/min free, 100/min Starter, 300/min Pro, 600/min Startup, 1000/min Enterprise.

### Which API has the best free tier?

Exa, with 20,000 free search requests a month. It is search only: contents, answers, and deep search cost credits. Tavily gives 1,000 credits (Researcher plan). Keiro gives 500 credits. SerpAPI gives 250 searches. Serper gives 2,500 credits one-time (not recurring). Firecrawl gives 1,000 credits (about 55 basic scrapes or 111 basic search calls, or 111 pages with JSON+Enhanced). Brave's free tier was removed in February 2026.

### Why is SerpAPI so expensive?

You are paying for 40+ search engines, 99.9% SLA, up to 100 QPS, and rich structured SERP (ads, knowledge panels, featured snippets), not for cheap queries. It is metadata only, so you also build the extraction layer. The rate drops to $9.17/1k at the Big Data tier (30k for $275), but at most volumes it is 10 to 15x a focused search API. The right pick when you need Google SERP verticals (shopping, maps, images) and can absorb the cost.

### Is per-credit pricing cheaper than per-query?

Sometimes, and that is the trap. Firecrawl at $0.83/1k credits looks cheaper than Keiro at $1/1k, but a Firecrawl search costs 2 credits per 10 results (rounded up), and a scrape with JSON+Enhanced costs 9 credits per page. Per-credit is cheaper when your workload matches the cheap unit (basic scrapes at 1 credit each) and more expensive when it does not (search calls at 2+ credits, JSON+Enhanced scrapes at 9 credits). Tavily at $0.005/credit on the Growth plan ($5/1k) is cheaper than its $0.008 PAYG rate ($8/1k), but requires a $500/mo commitment. Convert to your real operations before comparing.

### What is the true cost of ownership for an AI search API?

The API bill plus the infrastructure you build around it. For SERP-only APIs (Serper, Brave, SerpAPI), add a separate extractor ($1-5/1k) plus glue code. For RAG, add an embedding service ($0.10-0.50/1k) unless the API bundles embeddings (Keiro does, in `/search/content`). Add a caching layer (Firecrawl's cached results still cost 1 credit, so caching = speed not savings). Add retry and fallback logic for production reliability. A SERP-only API at $1/1k plus a $2/1k extractor plus a $0.30/1k embedding service is $3.30/1k all-in, plus engineering. Keiro at $3/1k (3 credits for `/search/content`) includes search, clean text, and embeddings in one call, no second service. The TCO gap is not just the API price.

### How do you compute the all-in RAG cost?

A RAG query is a search plus the clean page text the agent reads, plus optionally embeddings. For providers that return content in the call (Keiro, Tavily, Firecrawl, Exa, Perplexity), the all-in is their content-endpoint price. For providers that return raw SERP only (Serper, Brave, SerpAPI), the all-in adds a cheap extractor plus an embedding service, because without them you do not have clean text or embeddings. Keiro's `/search/content` is 3 credits, which at $1/1k credits is about $3/1k for search plus clean text plus optional embeddings, in one call. Tavily's basic search with inline content is 1 credit ($0.005-0.008 depending on plan). Firecrawl's search-plus-scrape is 2cr (search) + 1cr (scrape) per page, but with JSON+Enhanced it is 2cr + 9cr = 11cr per result page.

### How do I compare AI search API pricing accurately?

Convert every provider to the cost of 1,000 of your actual call type (basic search, search-plus-content, search-plus-content-plus-embeddings, answer, deep research, scrape). Add the per-operation credit multipliers (Firecrawl's 9cr/page, Tavily's 15-250cr research). Add the hidden costs (separate extraction, token fees, rate-limit upgrades, embedding service). Apply your real volume to the tier table (SerpAPI drops 2.7x at Big Data, Serper drops 3.3x at 12.5M, Tavily drops 1.6x on Growth). Model a full month. The headline per-1k rate alone will mislead you on Firecrawl (credit multipliers), Tavily (plan tiers), Perplexity (tokens), and SerpAPI (volume tiers) every time.