Blog

Tavily vs Exa (2026): Pricing, Search and Content

By Kartey R

Founder of Serpex

Updated 6 min read
Technology

Pick Tavily if you want a search call that can also hand back page content and a generated answer, with lots of request controls, on a credit plan. Pick Exa if your queries read like research questions and you want to choose how much work each search does, from a $4 instant search to a $15 deep-reasoning one. On plain search price they're close: Tavily is $5 to $8 per 1,000 and Exa is $4 to $7 for its standard modes.

If cost per search or page content in the same call decides it, there is a third option, and it's the one we make: Serpex, a search API for AI agents and LLM apps. One request returns up to 10 results and, with include_content, the page content of the top 5 or 10 as Markdown, at $0.50 to $0.80 per 1,000 searches, or $3.00 to $4.80 per 1,000 with 10 pages. Errors and timeouts are never billed, and new accounts get 200 free credits with no card.

Prices checked 1 October 2026 on Tavily's credits page and exa.ai/pricing.

How each one works

Tavily is a search API built for RAG pipelines. One search call returns ranked results, and each result's content is made of short relevant chunks of the page (up to 3 chunks per source, each up to 500 characters). Turn on include_raw_content and you also get the cleaned page as markdown or text. Turn on include_answer and Tavily adds an LLM-generated answer. Separate endpoints cover Extract (known URLs), Map, Crawl and Research.

Exa describes itself as a web search tool designed for agents. You pick a search type per request: Instant, Fast/Auto, Deep or Deep-Reasoning. Each costs more and does more work than the one before it. Page text comes from the Contents side of the API, priced per page per content type (text, highlights, summaries). Exa also sells an Answer endpoint, Monitors for recurring searches, and an Agent API for longer research runs.

Pricing side by side

SerpexTavilyExa
Standard search, per 1,000$0.80, $0.65 or $0.50 by top-up size; quoted from $0.20 at 10M+ a month$8.00 pay as you go, $7.50 to $5.00 on monthly plans$4 Instant, $7 Fast/Auto
Deeper search, per 1,000—Advanced depth is 2 credits, so $16.00 pay as you go$12 Deep, $15 Deep-Reasoning
Results includedUp to 10Up to 20 per searchUp to 10 at the base price, $1 per 1,000 requests for each extra result
Page contentinclude_content on the search call: 3 credits for 5 pages, 6 for 10, in totalinclude_raw_content on the search call; Tavily's credit table lists no separate charge for it$1 per 1,000 pages per content type
Extraction of known URLsExtract API turns any URL into Markdown1 credit per 5 successful URLs (basic), 2 per 5 (advanced)Contents endpoint, $1 per 1,000 pages per content type
Generated answer—include_answer on a searchAnswer endpoint, $5 per 1,000
Free tier200 free credits on signup, no card1,000 credits a month, no card$10 credit a month plus a $10 onboarding bonus, no payment method
BillingPay as you go; credits never expire; errors and timeouts never billedCredits: pay as you go at $0.008, or monthly plansPay as you go per request

Tavily's monthly plans, from its docs: Project is $30 for 4,000 credits ($0.0075 each), Bootstrap $100 for 15,000, Startup $220 for 38,000, and Growth $500 for 100,000 ($0.005 each). Credits reset on the first of each month.

So at small volume Exa's Instant search is the lower-priced of the two, at $4 per 1,000. At Tavily's Growth plan a basic search is $5 per 1,000, with page content in the same call. Once you add page text on Exa, the per-page charge comes on top of the search price.

Content handling

This is where the two feel most different.

With Tavily, one request can return everything a RAG step needs: results, relevant chunks, the full cleaned page and an answer. You choose markdown or text for the raw content. Extract failures aren't charged: "You never get charged if a URL extraction fails."

With Exa, content is its own priced unit. You pay for each page and each content type you ask for, so text plus a summary is two units per page. That's more to keep track of, and it also means you only pay for the content types you actually use.

Request controls

Tavily's search endpoint exposes a lot: search_depth (basic, fast, ultra-fast or advanced), topic (general, news or finance), time_range, start and end dates, include_domains and exclude_domains, country, max_results up to 20, and auto_parameters to let Tavily choose settings. Note that auto_parameters may switch a search to advanced depth, which costs 2 credits.

Exa's main control is the search type, which sets how much work a query gets. Check Exa's API reference for the current filter parameters before you rely on one.

SDKs and integrations

SerpexTavilyExa
Pythonserpex (PyPI)tavily-pythonexa-py
JavaScript / TypeScriptserpex (npm)@tavily/coreexa-js
MCPserpex-mcpRemote MCP server at mcp.tavily.comHosted MCP server at mcp.exa.ai
Frameworks listed in docsLangChain, LlamaIndex, Haystack and n8n integrationsLangChain, LlamaIndex, Vercel AI SDK, CrewAI and moreLangChain, LlamaIndex, Vercel AI SDK, OpenAI and Anthropic tool calling

Both are easy to wire into an agent. Tavily's docs list a longer set of framework integrations.

Pick Tavily if

  • You want results, page text and an answer from one call.
  • Your agent needs topic, date, domain or country controls as request parameters.
  • You want crawl and map endpoints from the same vendor.
  • Your usage fits inside 1,000 free credits a month.

Pick Exa if

  • Your queries are research questions written in natural language.
  • You want to choose per request between an instant search and a deep, reasoning one.
  • You only need page text for some results and would rather pay per page.
  • You'd rather pay per request than size a monthly plan.

A third option if cost per search is the main issue

If the deciding number is cost per search, Serpex is a search API for AI agents and LLM apps that costs $0.50 to $0.80 per 1,000 searches on pay-as-you-go credits that never expire, with no subscription. Adding include_content: true returns page markdown for the top 5 or 10 results in the same request, at 3 or 6 credits in total, billed on pages actually delivered. Errors and timeouts are never billed. New accounts get 200 free credits with no card.

It does less than either of them: no generated answers, no domain or date filter parameters (you write site: and after:/before: into the query instead), and no deep search modes. If you need those, stay with Tavily or Exa. The short versions of this comparison are on Tavily alternative and Exa alternative, and the wider price table is in Search API Pricing in 2026.

FAQ

Is Exa cheaper than Tavily?

For plain search at small volume, yes: Exa's Instant search is $4 per 1,000 and Fast/Auto is $7, against Tavily's $8 at pay as you go. On Tavily's $500 Growth plan a basic search drops to $5 per 1,000. Once you add page content, Exa charges $1 per 1,000 pages per content type, while Tavily's credit table lists no extra charge for raw content on a search.

Which has the better free tier?

Tavily gives 1,000 credits every month, which is 1,000 basic searches. Exa gives $10 in credit every month plus a $10 bonus at signup. That's $20 in the first month, enough for 5,000 Instant searches, then $10 a month, enough for 2,500.

Can either one write the answer for me?

Both can. Tavily adds an answer with include_answer on a search. Exa has a separate Answer endpoint at $5 per 1,000.

Try Serpex

Sign up for 200 free credits, no card required, and run the queries you're testing on Tavily and Exa. The docs show search with include_content, and pricing lists the rates by top-up, with credits that never expire.

Serpex Blog

Related Articles

More content you might be interested in

Oct 24, 2026 · 8 min read

Vibe Coding Web Search: Add a Search API With Cursor, Claude Code or Lovable

Add web search to an app you build with AI, no coding needed: get a key, give your agent the docs, keep the key server-side. Cursor, Claude Code, Lovable.

Technology
Read More
Oct 24, 2026 · 14 min read

Search APIs for Enterprise AI (2026): Security, Rate Limits and Cost at Scale

Questions to ask before signing a search API contract for enterprise AI: security, retention, SLAs, rate limits, billing, and cost at 1M-10M searches.

Technology
Read More
Oct 23, 2026 · 15 min read

PHP2Shell vs the Knowledge Cutoff: CVE-2026-109618, Tested Oct 2026

Only 3 of 36 top-3 search results for PHP2Shell came from the PHP project. We built a 91-line agent that checks CVE-2026-109618 against primary sources.

Technology
Read More
Oct 19, 2026 · 19 min read

Which Search API Is Cheapest at 100k, 1M and 10M Searches a Month? (Prices Checked October 2026)

Search API costs at 100k, 1M and 10M searches a month, checked Oct 23, 2026. Lowest list: $0.20/1k. Serpex: $0.50-$0.80, from $0.20 at 10M+.

Technology
Read More
Oct 19, 2026 · 8 min read

How to Add Web Search to an AI Workflow or RAG Pipeline (Python, 2026)

How to add web search to an AI workflow or RAG pipeline in Python (2026): one call returns pages as Markdown to chunk, rank and cite, with real monthly costs.

Technology
Read More
Oct 19, 2026 · 8 min read

Tavily Pricing in 2026: What 10k, 100k and 1M Searches Cost

Tavily pricing in 2026: $8 per 1,000 credits pay as you go, $5 on Growth. Monthly cost at 10k, 100k and 1M searches, advanced depth, Extract and alternatives.

Technology
Read More
Oct 19, 2026 · 8 min read

Serper vs Tavily (2026): Pricing, Output and Which to Use

Serper vs Tavily in 2026: Serper costs $0.30 to $1.00 per 1,000 searches, Tavily $5 to $8. Free tiers, output, page content, LangChain support and which to use.

Technology
Read More
Oct 19, 2026 · 8 min read

Google Custom Search API Is Closing: What to Use by January 1, 2027

The Google Custom Search JSON API is closed to new users and existing ones must move by January 1, 2027. Replacements and 2026 costs at 10k and 100k searches.

Technology
Read More
Oct 3, 2026 · 7 min read

Web Search in n8n: HTTP Request, AI Agent or Community Node

Add web search to an n8n workflow with the HTTP Request node, use it as an AI Agent tool with $fromAI(), or install the n8n-nodes-serpex community node.

Technology
Read More