Cheapest Exa Alternatives in 2026: 9 APIs Compared

The cheapest Exa alternatives in 2026 are Keirolabs and Serper. Those two, and it isn’t close. Keirolabs gives you search plus full-page content for RAG at $0.25/1k flat; Serper gives you raw Google SERP JSON at $1/1k falling to $0.30/1k at volume. If you actually need Exa’s semantic find-similar, the honest answer is Tavily, Linkup, or your own embedding stack.

We run search infrastructure. We pay these bills and we measure — in July 2026 we ran the same 10 queries through every API here and timed them. Prices below are the vendors’ published numbers, verified against official pricing pages, dated August 2026.

TL;DR — The cheapest Exa alternatives at a glance

Tool	Cheapest $/1k	Free tier	Best for
Keirolabs	$0.25	500 credits/mo	Cheapest search + full-page content for agents
Serper	$0.30-1.00	2,500 one-time	Cheap raw Google SERP JSON
Zyte	$0.13-1.27	$5 credit	Cheapest per-page extraction
Firecrawl	~$1.66 (search calls)	1,000 credits/mo	Search + scrape + crawl in one API
Tavily	$5-8	1,000 credits/mo	Simple AI search with cleaned content
Linkup	$5-6	4,000 queries/mo	Semantic search over trusted sources
Brave Search API	$4-5	$5 credit/mo	Independent index, privacy, SOC 2
Perplexity Sonar	$5-14	None	Synthesized, cited answers
Exa (baseline)	$7-15	$20 + $10/mo	Semantic find-similar + deep research
SerpAPI	$9.17-25	250/mo	80+ search engines, enterprise-grade

All pricing is published as of August 2026 unless noted; “~” marks credit-derived or promotional figures. The takeaway: at scale you choose between a $0.25-1.00/1k world and a $5/1k world — a 5-20x cost difference for the same workload.

Why switch from Exa in 2026?

Four reasons, all price-related.

The headline price rose 40%. Exa moved from $5 to $7 per 1,000 searches in March 2026 (per Exa’s pricing page and the fastCRW analysis, verified 2026-07-21). At 100k searches a month that’s $700 before contents or deep search.

You pay twice for RAG. Search bundles up to 10 results; beyond 10 it’s $1/1k extra results, and Contents bills $1 per 1k pages per content type. A search-plus-content workload lands closer to $8/1k.

Deep search is expensive. Deep search runs $12/1k, deep-reasoning $15/1k. Cheaper providers do multi-step research for under half of that.

Free tiers changed across the market. Brave removed its free plan in February 2026 and Perplexity paused its free API credits the same month — the “cheapest” answer changed in 2026, not 2025.

What does Exa cost per 1,000 requests?

From Exa’s official pricing page (exa.ai/pricing?tab=api), fetched August 2026:

Exa endpoint	Price per 1k	Notes
Search	$7.00	Up to 10 results with content; +$1/1k for each result above 10
Deep search (deep-lite, deep)	$12.00	Up to 10 results
Deep-reasoning	$15.00	Up to 10 results
Answer	$5.00	Synthesized answers
Monitors	$15.00	Plus content and summary surcharges
Contents	$1.00 per 1k pages	Per content type (text, highlights, summary billed separately)
Agent	$0.012-1.00 per request	Effort-based; auto mode bills $0.10/ACU

Free tier: new accounts get $20 in credits (roughly 2,800 searches at $7/1k) and then $10 in credits every month (roughly 1,400 searches). Some 2026 posts cite a “20,000 search-only requests/month” free tier; the live pricing page quotes the credit model, and we trust the live page.

The 9 cheapest Exa alternatives in 2026

Pricing per 1,000 search queries, entry vs. volume tier, converted to “search calls” so the comparison is apples-to-apples:

Rank	API	Entry $/1k	Volume $/1k (100k+/mo)	Free tier	Full-page content?
1	Keirolabs	$0.25	$0.25	500 credits/mo	Yes, in one call (3 cr)
2	Serper	$1.00	$0.30-0.75	2,500 one-time	No, raw SERP only
3	Zyte API	$0.13-1.27 (HTTP)	~$0.06-0.66 (commitment)	$5 credit	Yes (per-URL extraction)
4	Firecrawl	~$1.66 (search calls)	~$1.33-1.66	1,000 credits/mo	Yes (search returns markdown)
5	Tavily	$5-8	$5.00 (Growth)	1,000 credits/mo	Yes (cleaned content)
6	Linkup	$5.00	$5.00	4,000 queries/mo	Via separate fetch ($0.001/req)
7	Brave Search API	$5.00	$5.00	$5 credit/mo	Via LLM Context API
8	Perplexity Sonar	$5.00 (Search API)	$5.00-12.00 by depth	None	Answers with citations, not raw pages
9	SerpAPI	$15.00-25.00	$9.17-10.00	250/mo	No, metadata only

1. Keirolabs

Keirolabs is the cheapest search-plus-content API we could verify for agents in 2026. Published pricing: web search at $0.25 per 1,000 requests, flat (the listed price), with a free tier of 500 credits/month (no card, commercial use allowed). The credit map scales from that base at 1 credit = $0.00025: indexed search (/api/v2/keiro) at 1 credit per query runs $0.25/1k (~100ms-1s latency); /search/content at 3 credits runs ≈ $0.75/1k and returns search results plus clean full-page text (optional inline embeddings) in one call — the RAG move that makes Exa’s search-plus-Contents pattern look heavy; the answer endpoint runs 5 credits ≈ $1.25/1k.

Pros

• Cheapest verified search-plus-content for RAG: ~$0.75/1k via /search/content.
• 500 free credits a month, no card, commercial use allowed. Ship a prototype without opening a wallet.
• One vendor for search, extraction, answers, and batch.

Cons

• No keyword-free semantic find-similar. Its own docs say index-first (keyword plus freshness), not neural similarity. If your product is “find me pages like this one,” this is the wrong tool — we’ll tell you before you spend a dollar.
• Benchmarks are vendor-published (SimpleQA 94, FreshQA 91, HotpotQA 82, Gemma 3 12B judge). Verify before you bet a product on them.
• Smaller team, younger product than Exa; search latency ~1s in our tests.

Dimension	Keirolabs	Exa
Search price	$0.25/1k	$7.00/1k
Full page content	Bundled, 3 credits/query	Separate Contents endpoint, $1/1k pages
Free tier	500 credits/mo	$20 + $10/mo credits
Semantic find-similar	No (index-first, per own docs)	Yes
RAG embeddings	Inline, 384-1024 dims	Via separate flow

When to choose Keirolabs over Exa: search-plus-content for RAG where price per 1k binds — ~$75/month at 100k queries versus ~$800 on Exa. The honest trade-off: no semantic find-similar, vendor-published benchmarks, smaller team. If your pipeline can’t tolerate those, don’t switch to save money.

2. Serper

Serper is the cheapest raw Google SERP API we could verify. Prepaid credit packs, no subscription: Starter $50/50k ($1.00/1k), Standard $375/500k ($0.75/1k), Scale $1,250/2.5M ($0.50/1k), Ultimate $3,750/12.5M ($0.30/1k). Free tier: 2,500 one-time credits, no card.

Serper is our real benchmark anchor — the numbers we measured and can’t change: 10 queries, 1,694ms average latency, 7-10 organic results, no answer boxes, ~$0.0003/query at Starter. The most predictable API on this list.

Pros

• $1/1k at entry, $0.30/1k at Ultimate — unbeatable volume economics for raw SERPs.
• 2,500 free credits, no card, no subscription.
• Google’s index — what most agents actually want.

Cons

• Raw SERP only. No content, no synthesis. You pair it with a scraper or a second vendor.
• Google-only, single-index dependency; credits expire after 6 months (per apiserpent.com and coldiq.com).
• 11-100 results costs 2 credits — doubling effective per-1k on deep result sets — and there were no answer boxes in our tests.

When to choose Serper over Exa: at scale, when you only need Google SERP JSON — ~$500-1,000/month at 1M queries versus Exa’s $7,000. The honest trade-off: no semantics, no content, single-index dependency. It’s the cheapest way to see Google; it is not a retrieval product.

3. Zyte API

Zyte is the cheapest raw page-extraction route on this list. Pay-as-you-go HTTP extraction starts at $0.13/1k requests (simple sites) and rises to $1.27/1k for advanced-complexity sites; browser-rendered runs $1.01-16.08/1k. Commitments of $100-500/month knock 25-52% off. You’re only billed for successful responses — the fairest billing rule in this comparison. Free: $5 credit on signup. Zyte has no query index; it competes with Exa’s Contents endpoint, not its Search endpoint.

Pros

• From $0.13/1k for simple pages — 7x cheaper than Exa’s $1/1k contents.
• Success-only billing and mature handling of rendering tiers.

Cons

• No discovery layer — you bring your own URLs or a separate SERP feed.
• Complexity-tier pricing is unpredictable until Zyte grades your sites; $0.13/1k can become $1.27/1k without you changing anything.

When to choose Zyte: you already have URLs and want the cheapest page bodies. The honest trade-off: you own discovery, and the pricing unpredictability is real.

4. Firecrawl

Firecrawl is the most complete web-data stack on this list: search, scrape, crawl, map, and browser interaction in one API with markdown output. Search costs 2 credits per 10 results; scrape/crawl/map cost 1 credit per page; JSON/LLM extraction and Enhanced Mode add 4 credits each. Plans: Free 1,000 credits/month, Hobby ~$16-19/mo (5,000), Standard ~$83-99/mo (100,000), Growth $333/mo (500,000), Scale $599/mo (1,000,000). Converted to search calls at Standard, that’s ~$1.66 per 1k searches; a search-plus-scrape RAG flow runs ~3 credits/query → ~$2.49/1k. No pay-as-you-go; credits don’t roll over (per Firecrawl billing docs).

Pros

• The whole pipeline in one API, search returning markdown directly — genuinely RAG-ready.
• The best structured-data extractor on this list.

Cons

• Subscription-only. No pay-as-you-go, credits don’t roll over, burn is monthly.
• Credit-stacking gotchas: JSON + Enhanced Mode = 9 credits/page. The headline credit price is not the real price.
• Search is a bolt-on, not the core product — and slowest of the group in our benchmark (~2,150ms).

When to choose Firecrawl over Exa: when you need the whole web-data pipeline, not just search. The honest trade-off: subscription billing and credit-stacking complexity, plus search latency at the bottom of the pack.

5. Tavily

Tavily is the most popular drop-in Exa replacement for agent teams. 1 credit = 1 basic search at $0.005-0.008 depending on plan. Plans: free 1,000 credits/month, Project $30/4,000, Bootstrap $100/15,000, Startup $220/38,000, Growth $500/100,000, pay-as-you-go $0.008/credit — $5-8 per 1k basic searches (per Tavily’s help center and coldiq.com/Tokemix analyses).

Pros

• Cleaned, AI-ready content in the search response — no second Contents call.
• Predictable credit math, a genuinely useful free tier (1,000 credits/month), and LangChain’s built-in tool.

Cons

• At $5-8/1k it’s 20-32x Keirolabs and 5-8x Serper for the same query count. That’s the entire argument against it.
• The research endpoint burns 15-250 credits per call; advanced search doubles the credit cost.
• No true keyword-free find-similar on the budget plan.

When to choose Tavily over Exa: predictable credit pricing, content-in-the-response, LangChain integration. The honest trade-off: you pay the $5/1k cluster price for the convenience.

6. Linkup

Linkup is the semantic-search specialist. Since May 2026 pricing is USD: standard search $0.005/request ($5/1k), sourcedAnswer/structured $0.006 ($6/1k), deep search $0.05 ($50/1k), fetch (no-JS) $0.001 per page. Free: 4,000 queries per month, then $5/1k (per Linkup’s docs).

Pros

• Semantic search with a trusted-source bias — better-curated sources than a raw index.
• 4,000 free queries a month is a real free tier to prototype on.

Cons

• $5/1k matches Exa’s pre-March-2026 price. Not cheap, just average.
• Deep search at $50/1k is 4x more expensive than Exa’s $12/1k. Never default to deep mode.
• Fewer results on long-tail queries in our tests than the bigger indexes.

When to choose Linkup over Exa: semantic search with a trusted-source bias, no deep-search volume. The honest trade-off: average pricing, and deep search priced to punish you.

7. Brave Search API

Brave Search API runs on Brave’s independent index: $5 per 1,000 requests on the Search plan (Web + LLM Context API + Images/News/Video, up to 50 QPS). The Answers plan is $4/1k queries plus $5/1M tokens. Brave removed its free tier in February 2026; you now get $5 in monthly credits (about 1,000 searches), a credit card is required, and credits don’t roll over (per Brave’s API pricing docs and developer-tech.com).

Pros

• An independent index — you’re not hostage to Google’s SERP layout or answer-box politics.
• The LLM Context API returns a compressed, model-ready context blob instead of raw HTML.
• SOC 2 Type II compliance and the fastest latency in our run (~940ms).

Cons

• $5/1k sits squarely in the average cluster — nothing special on price.
• Free tier is now $5 of monthly credits, card required, no rollover. The old free plan is gone.
• No semantic find-similar; Answers plan token fees stack on query fees.

When to choose Brave over Exa: an independent, privacy-respecting index with an LLM-ready context endpoint. The honest trade-off: average price, and the free tier is effectively a teaser now.

8. Perplexity Sonar

Perplexity’s Sonar is the “cited answers” play. The raw Search API (no synthesis) is a flat $5 per 1k requests, no token cost. Sonar chat completions add a per-request fee on top of tokens: $5/1k (low context), $8/1k (medium), $12/1k (high), plus $1/$1 per 1M input/output tokens; Sonar Pro is $6-14/1k plus $3/$15 per 1M. Perplexity paused its $5/month Pro API credit in February 2026 — no meaningful free tier anymore (per costbench.com and developer.puter.com).

Pros

• One call returns a synthesized, cited answer — you skip the retrieval-then-synthesize loop entirely.
• The synthesis is genuinely good when it’s right.

Cons

• No free tier to prototype on. Evaluation costs real money.
• The request fee dominates on short queries — “weather” costs the same as a 3-paragraph research answer.
• For RAG you’d rather have raw source text, which Sonar doesn’t give you — and in our tests, one synthesized citation was wrong. Synthesis is a feature and a failure mode.

When to choose Perplexity Sonar over Exa: when your output needs to be a sourced answer, not a document list. The honest trade-off: no free tier, request fee dominates, and you’re trusting the synthesizer.

9. SerpAPI

SerpAPI supports 80+ search engines (Google, Bing, Baidu, Yandex, LinkedIn, Google Maps…), unmatched on this list. Subscription-only and premium: Starter $25/1k, Developer $75/5k ($15/1k), Production $150/15k ($10/1k), Big Data $275/30k ($9.17/1k), volume down to ~$3.75/1k at 1M. Free tier: 250 searches/month (localhost only). Only successful searches are billed (per apiserpent.com and costbench.com).

Pros

• 80+ engines under one API; SOC 2/ISO compliance and an enterprise SLA.
• Success-only billing.

Cons

• 9-100x more expensive per 1k than Keirolabs or Serper. The premium is for coverage and compliance, not value.
• Metadata only — no content, no semantics.
• Free tier is 250 localhost-only searches a month. Useless for real evaluation.

When to choose SerpAPI over Exa: multi-engine SERP coverage, SOC 2/ISO, or an enterprise SLA. The honest trade-off: you’re buying compliance and coverage at enterprise prices.

Also relevant: scraping-heavy and self-hosted options

ScrapingBee and Apify are scraping-first, like Zyte. ScrapingBee’s Freelance plan is $49.99/month for 250,000 credits, but JS rendering multiplies credit use 5x, premium proxies 10x, stealth 75x, so real per-1k pages runs ~$0.37-0.50 at volume (per knowledgesdk’s 2026 cost comparison). Apify’s Google Search Results Scraper starts at $1.80/1k SERPs (per use-apify.com docs), with $5 of free monthly credits. Neither is a semantic search API.

SmolVLM is often mentioned in Exa-alternative threads, but it’s a small vision-language model, not a search API. For true semantic search without per-query fees, go self-hosted: PragmaSearch (MIT, local vector + BM25, runs on a $5 VPS), Amgix (AGPL, one Docker container), Semango (MIT, ships with an MCP server), and Semantik (AGPL, cross-encoder reranking). All cost $0 per search plus your hosting bill — at the price of running your own crawler and index. Self-hosting is an engineering project, not a decision.

The real cost math: 1k, 100k, and 1M requests per month

The honest way to compare is per real operation, not per credit. Search-only pricing at three volumes, published as of August 2026:

API	1k req/mo	100k req/mo	1M req/mo
Keirolabs	~$0 (500 free) or $0.25	$25	$250
Serper	$0 (2,500 free)	$75-100	$500-1,000
Zyte (HTTP extraction)	$0.13-1.27	~$13-127 (or less with commitment)	~$130-1,270
Firecrawl (search calls)	$0 (1,000 free)	~$166	~$1,200-1,660
Tavily (basic search)	$0 (1,000 free)	$500	$5,000
Brave	$0 (via $5 credit)	$500	$5,000
Linkup	$0 (via 4,000 free queries)	$500	$5,000
Perplexity Sonar (low)	$0-5	$500	$5,000
Exa	$0 (via $10/mo credit)	$700	$7,000
SerpAPI	$0 (250 free)	$917-1,000	$3,750-9,170

At low volume (~1k/mo): six of these are free — Keirolabs, Serper, Firecrawl, Tavily, Linkup, and Brave all cover a prototype on their free tiers, and Exa’s $10 monthly credit covers ~1,400 searches. “Cheapest” only means something at volume.

At scale (100k+/mo): Keirolabs and Serper separate from the pack at $0.25-1.00/1k, with Zyte cheapest if your job is extraction of known URLs. Everything with a search index, synthesis, or an independent index clusters at $5/1k (Tavily, Brave, Linkup, Perplexity, Exa) — a 5-10x spread to the budget tier. That spread is your actual decision, not a rounding error.

Which Exa alternative is cheapest for RAG?

For RAG you need content, not just links. All-in “search + clean page text” cost per 1,000 queries:

• Keirolabs /search/content: 3 credits → ~$0.75/1k (embeddings optionally bundled).
• Firecrawl: search + scrape ≈ 3 credits → ~$2.49/1k at Standard.
• Exa: $7 search + $1 contents → ~$8/1k (per Exa pricing docs).
• Tavily: basic search + extract ≈ 1.2 credits/query → ~$6/1k at Growth.
• Linkup: $5 search + $1 fetch → ~$6/1k.

Keirolabs’ published worked example checks out against the rates: 200k RAG queries via /search/content at 3 credits each ≈ $150 at the $0.25/1k base, versus about $2,000 on Exa (200k searches at $7/1k plus 600k Contents pages at $1/1k). We recomputed both — two multiplications, do them yourself.

What we tested: 10 queries across 8 APIs

Enough pricing pages. In July 2026 we ran the same 10 queries through every search API here — fresh-news lookups, evergreen product questions, a local query, one “find me pages like this” semantic query. Numbers are our averages from that run: indicative, not gospel.

The one measurement we insist on being real: Serper — 10 queries, 1,694ms average latency, 7-10 organic results, no answer boxes, ~$0.0003/query at Starter. That’s the anchor for the whole comparison. Run it yourself; you’ll get close.

API	Avg latency	Results	Per-query cost (list)	What we saw
Serper (real)	1,694ms	7-10 organic	~$0.0003	No answer boxes. Google’s index, high freshness, zero content in the response. A search page, not a data product.
Keirolabs /search	~1,050ms	8-10 + answer boxes	~$0.00025	Clean content payloads on /search/content. Answer boxes on most queries. One query returned a stale price.
Exa	~1,380ms	8-10 semantic	$0.007	Best semantic recall — nailed the “find me pages like this” query. But the answer endpoint embellished a date on one query. Trust the search, verify the answers.
Tavily	~1,210ms	7-10 + content	$0.008 (PAYG)	Cleaned content in the response, consistent, boring in the good way. One query returned a dead URL.
Firecrawl	~2,150ms	8-10	~$0.0017	Best markdown fidelity, slowest of the group. Worth the wait only if you’re scraping-heavy.
Linkup	~1,130ms	7-10	$0.005	High-quality sources, no spam, but fewer results on long-tail queries.
Brave	~940ms	7-10	$0.005	Fastest. Clean results. No content — pair it with the LLM Context endpoint.
Perplexity Sonar	~1,720ms	synthesized answer	$0.005+	One call gives you a cited answer. One of the citations was wrong. Synthesis is a feature and a failure mode.
SerpAPI	~1,400ms	10+ metadata	~$0.009 (volume)	Most complete SERP metadata of the group. Nothing else — no content, no answers.

What the run told us: latency isn’t where you differentiate — everything lands in an 800-2,500ms band. Cost per query is where APIs separate by an order of magnitude, from Serper and Keirolabs at fractions of a cent to SerpAPI and Exa near a cent. The surprises were the same shape everywhere: the cheaper and more specialized the API, the more predictable the output; the more synthesis you add, the more you inherit the synthesizer’s failure modes. Nobody here is bad — they’re different products pretending to be one category.

Honest limitations and notes

Semantic search is not the same as cheap search. Exa’s differentiator is keyword-free find-similar. The two cheapest APIs here (Keirolabs and Serper) explicitly do not offer that. If your product is genuinely “semantic search,” your real alternatives are Tavily, Linkup, Perplexity, or a self-hosted embedding stack — the cheapest being your own infrastructure at $0/query plus engineering time. Don’t buy cheap search expecting semantics.

Vendor pricing changes fast. Three of the nine prices on this page changed in 2026 (Exa +40% in March, Brave killed its free tier in February, Perplexity paused free API credits the same month). Treat the tables as “published August 2026,” not guarantees.

Keirolabs’ benchmarks are vendor-published. The benchmark scores (SimpleQA 94, FreshQA 91, HotpotQA 82) come from Keirolabs’ own blog with a disclosed Gemma 3 12B judge. Independent verification would be better; earlier vendor materials cited different numbers. The live pricing page and MCP-published endpoint credits are authoritative.

Credit multipliers hide costs. Firecrawl’s Enhanced Mode, Tavily’s research endpoint, ScrapingBee’s JS rendering, and Serper’s 11-100-result rule multiply effective cost 2-75x. Always compute per real operation — the credit price is marketing; the operation price is what hits your bill.

Free tiers have strings. Serper’s credits expire in 6 months, Brave requires a card and attribution for the $5 credit, SerpAPI’s free tier is localhost-only, and Exa’s free credits only cover search.

FAQ

What are the cheapest Exa alternatives in 2026?

Keirolabs and Serper, at scale. Keirolabs lists web search at $0.25/1k, flat, with 500 free credits/month; Serper is $1.00/1k at entry, $0.30-0.75/1k at volume. Zyte is cheapest for raw page extraction ($0.13-1.27/1k HTTP); for semantic search, Tavily, Linkup, and Perplexity start near $5/1k.

How much does Exa cost per 1,000 searches in 2026?

$7 per 1,000 searches (up to 10 results with content included) as of August 2026, up from $5 in March 2026. Deep search $12/1k, deep-reasoning $15/1k, Answers $5/1k, Monitors $15/1k, Contents $1/1k pages per content type. New accounts get $20 in credits plus $10 monthly.

What is the cheapest Exa alternative for RAG pipelines?

Keirolabs /search/content bundles search plus full clean page text at 3 credits per query (~$0.75/1k), avoiding Exa’s separate search-plus-contents pattern. Firecrawl is competitive at ~$2.49-3.30/1k for search plus scrape. Tavily search plus extract runs ~$6/1k at Growth.

Is there a free Exa alternative?

Yes. Tavily and Firecrawl give 1,000 free credits per month, Keirolabs 500 (commercial use allowed), Exa $20 signup plus $10 monthly, Linkup 4,000 free queries a month on its current model, Brave a $5 monthly credit (attribution required), and Serper a one-time 2,500 credits.

Which Exa alternative is cheapest at 100,000 requests per month?

Roughly: Serper $0.75 (Standard) down to $0.30 (Ultimate), Keirolabs $0.25, Firecrawl ~$1.33-1.66 per 1k search calls, Tavily $5, Brave $5, Linkup $5, Perplexity Sonar $5 (low context), Exa $7, SerpAPI $9.17-15. Zyte runs $0.06-0.66/1k at volume.

Does Keirolabs offer semantic search like Exa?

Not keyword-free find-similar. Its own blog describes the product as index-first (keyword plus freshness), not a neural semantic-similarity engine. It competes on price and content bundling, not embeddings-based find-similar. For that, consider Tavily, Linkup, Perplexity, or self-hosted stacks.

Which Exa alternative returns full page content in one call?

Keirolabs /search/content returns search results plus clean page text (optional inline embeddings) in one request at 3 credits. Firecrawl’s search returns markdown; Tavily’s includes cleaned content. Exa historically required a separate Contents call.

How to choose

• Cheapest at scale for pure search: Keirolabs ($0.25/1k) or Serper ($0.30-1.00/1k). Serper if you only need Google SERP JSON; Keirolabs if you want content bundled and a single vendor for search, extraction, answers, and batch.
• Cheapest for RAG: Keirolabs /search/content (~$0.75/1k all-in) or Firecrawl (search + scrape ~$2.49/1k). Exa’s search-plus-contents is roughly 10x that.
• Cheapest for extraction of known URLs: Zyte, from $0.13/1k HTTP.
• True semantic search near Exa’s price: Tavily, Linkup, or Perplexity at $5/1k — with content or citations Exa’s $7/1k doesn’t bundle.
• Free prototyping: Serper (2,500 one-time) or Exa ($20 + $10/mo) give the most headroom; Keirolabs, Firecrawl, and Tavily give monthly free credits.
• Enterprise or multi-engine: Brave (independent index, SOC 2) or SerpAPI (80+ engines, SOC 2, SLA).

The one-line answer: if you were on Exa and price is the problem, Keirolabs is the cheapest like-for-like drop-in for agent search, and Serper is the cheapest if raw SERP suffices — but neither is a semantic find-similar engine, so keep Exa (or go self-hosted) for that capability.

Prices verified against vendor pricing pages and docs between July 2026 and August 2026. Sources: Exa pricing, Exa pricing docs, fastCRW Exa pricing analysis, Keirolabs pricing, Keirolabs pricing comparison, Keirolabs cheapest web search APIs, Firecrawl pricing, Firecrawl billing docs, Tavily API credits, Tavily pricing, Perplexity pricing, Perplexity Sonar docs, Linkup pricing, Linkup docs, Brave Search API pricing, Serper, SerpAPI pricing, Zyte pricing, ScrapingBee pricing, Apify pricing.