---
type: analysis
slug: guide-search-apis-for-agents
title: "Search APIs for agents: Exa, Parallel, Tavily and the rest"
description: "Search APIs for agents return web results shaped for LLMs instead of people: ranked URLs with extracted text, highlights or compressed excerpts, and in some products cited answers or asynchronous deep-research runs. Exa, Parallel, Tavily, Brave Search API, Perplexity's Search and Agent APIs and Fire"
published_at: 2026-10-28T09:00:00.000Z
topics: ["tools","developer_tools"]
source_urls: ["https://exa.ai/pricing","https://exa.ai/docs/reference/search","https://parallel.ai/pricing","https://docs.tavily.com/documentation/api-credits","https://brave.com/search/api/","https://docs.perplexity.ai/docs/getting-started/pricing","https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview","https://www.firecrawl.dev/pricing"]
author: "Andrey Reshetnikov"
author_profile: https://newruntime.com/owner-profile.md
---
Search APIs for agents return web results shaped for LLMs instead of people: ranked URLs with extracted text, highlights or compressed excerpts, and in some products cited answers or asynchronous deep-research runs. Exa, Parallel, Tavily, Brave Search API, Perplexity's Search and Agent APIs and Firecrawl search differ mainly in latency mode (instant or fast versus deep), what comes back (raw results, page contents, answers), billing unit (per 1k requests, per credit, per run) and free allowance. This brief is for engineers choosing a search tool for coding agents, chat assistants or research subagents on cost, latency and output shape.

*Field guide compiled 2026-10-23 from the official documentation, pricing pages and changelogs linked below; every line links to the page that states it.*

## What it does

- **Search.** Exa web search tool calls for agents returning webpage text and highlights, with configurable latency of 450ms to 1s. How: exa.search("query", type="auto", contents={"highlights": True}) [source](https://exa.ai/pricing)
- **Deep Search.** Exa multi-step agent workflows that return answers with structured outputs and web-grounded citations. How: Exa search with a deep type (deep-lite, deep, deep-reasoning); additionalQueries works only with deep-search types [source](https://exa.ai/docs/reference/search)
- **Search API.** Parallel returns ranked web URLs with token-dense compressed excerpts for agent tool calls, 200ms to 3s synchronous, 600 requests/min. How: Parallel Search API with a search objective and keywords [source](https://parallel.ai/pricing)
- **Task API.** Parallel deep web research and structured enrichments with citations, reasoning and confidence, asynchronous 5s to 30min. How: Parallel Task API with structured data and a question [source](https://parallel.ai/pricing)
- **Tavily Search.** Tavily search whose cost depends on search depth: basic costs 1 API credit and advanced costs 2 API credits per request. How: searchdepth basic or advanced [source](https://docs.tavily.com/documentation/api-credits)
- **LLM Context.** Brave endpoint returning ranked, extracted passages for models in one call, configurable from 1k to about 33k tokens with a relevance threshold. How: Brave Search API LLM Context endpoint with contextthresholdmode [source](https://brave.com/search/api/)
- **Goggles.** Brave custom re-ranking and domain sets for results. How: Pass a Goggle with Brave Search API requests [source](https://brave.com/search/api/)
- **Search API with Fast Search.** Perplexity lower-latency raw web search results; a request can carry up to five queries and is billed once. How: POST /search with searchtype: "fast" [source](https://docs.perplexity.ai/docs/getting-started/pricing)
- **Agent API.** Perplexity's replacement for Sonar Chat Completions: one API for frontier and open models with built-in web search, URL fetching, sandboxes and MCP, and effort levels fast, low, medium, high, xhigh. How: Migrate Sonar Chat Completions calls to the Agent API (Open Responses compatible) [source](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview)
- **Monitors.** Exa runs searches on a cadence to identify fresh events on the web. How: Exa Monitors API [source](https://exa.ai/pricing)

## Modes

- **auto.** Exa default balanced search mode; other types are lite, instant, fast, deep-lite, deep and deep-reasoning. [source](https://exa.ai/docs/reference/search)
- **instant.** Exa lowest-latency search type for latency-sensitive tool calls. [source](https://exa.ai/docs/reference/search)
- **deep-reasoning.** Exa deepest search type for multi-step, reasoning-heavy queries. [source](https://exa.ai/docs/reference/search)
- **Responses API.** Parallel OpenAI-compatible cited answers from the live web in 5s to 60s, $0.01 - $0.25 per request. [source](https://parallel.ai/pricing)
- **advanced.** Tavily search depth that costs 2 credits instead of 1 for basic. [source](https://docs.tavily.com/documentation/api-credits)
- **fast.** Perplexity searchtype for lower-latency results at $1.00 per 1K requests instead of $5.00. [source](https://docs.perplexity.ai/docs/getting-started/pricing)
- **websearch (Fast Search).** Perplexity Agent API tool at $0.001 per invocation; standard websearch is $0.0025 per invocation. [source](https://docs.perplexity.ai/docs/getting-started/pricing)

## Pricing and limits

- **Exa Search:** From $4 /1k requests. Starter free: $10 credits every month plus a $10 onboarding bonus, 10 Search QPS; Developer pay as you go up to 25 Search QPS [source](https://exa.ai/pricing)
- **Exa Deep Search:** $12 /1k requests. Contents $1 /1k pages per content type; Agent $0.012–$1.00 /run [source](https://exa.ai/pricing)
- **Parallel Search API:** $0.001 - $0.005 for 10 results per request. Up to 5,000 requests per month free; $5 in free credits per month; 600 requests/min [source](https://parallel.ai/pricing)
- **Parallel Task API:** $0.005 - $2.4 per request. 2,000 requests/min [source](https://parallel.ai/pricing)
- **Tavily Pay As You Go:** $0.008 / credit. Monthly plans $0.0075 - $0.005 per credit, e.g. Project 4,000 credits for $30 [source](https://docs.tavily.com/documentation/api-credits)
- **Brave Search API Search:** $5 per 1,000 requests. Includes $5 in free credits every month; 50 queries per second [source](https://brave.com/search/api/)
- **Brave Search API Answers:** $4 per 1,000 requests + $5 per million input/output tokens. Includes $5 in free credits every month [source](https://brave.com/search/api/)
- **Perplexity Search API:** $5.00 per 1K requests; $1.00 per 1K with Fast Search. Invalid, rate-limited and upstream-failed requests are not billed; no token charges [source](https://docs.perplexity.ai/docs/getting-started/pricing)
- **Firecrawl Free Plan:** $0 /month. 1,000 credits / month, i.e. 500 searches or 1,000 pages scraped, 2 concurrent requests; Hobby $16/month billed yearly for 5,000 credits [source](https://www.firecrawl.dev/pricing)

## Recent changes

- 2026-09-27: Perplexity ended Sonar Chat Completions support and recommends the Agent API for all new projects. [source](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview)

## Alternatives

- **Firecrawl.** Credit-based search and scraping in one plan: a search costs 2 credits (500 searches or 1,000 scrapes on the free 1,000 credits), so it suits agents that search and then scrape full pages. [source](https://www.firecrawl.dev/pricing)
- **Brave Search API.** Own independent index with no Google or Bing fallback and prepaid credits; Brave reports 669 ms average latency in an AIMultiple test versus 998 ms for Tavily and 1,200 ms for Exa (vendor-published comparison). [source](https://brave.com/search/api/)

## Traps

- Perplexity Sonar Chat Completions support ended on September 27, 2026; synchronous and streaming calls are being reformulated as Agent API requests, but asynchronous Sonar requests are no longer supported and must move to background mode. [source](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview)
- Perplexity Search API bills a successful request even when it returns no results. [source](https://docs.perplexity.ai/docs/getting-started/pricing)
- On Tavily, once monthly credits are exhausted requests stop until credits reset on the first of the month or you upgrade; advanced search burns credits twice as fast as basic. [source](https://docs.tavily.com/documentation/api-credits)
- Parallel Find All API is limited to 25 requests per hour and takes 10min to 1hr asynchronously, so it is not a live tool-call search. [source](https://parallel.ai/pricing)
- Firecrawl pay-as-you-go top-ups are for paid plans only; the free plan has low rate limits and 2 concurrent requests. [source](https://www.firecrawl.dev/pricing)
- Benchmark charts on Brave's page are Brave's own published runs or vendor-selected independent tests; check them against your own queries before choosing on accuracy claims. [source](https://brave.com/search/api/)

Author: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Tools](https://newruntime.com/topics/tools.md): Explore the tools topic hub.
- topic hub: [Developer Tools](https://newruntime.com/topics/developer-tools.md): Explore the developer_tools topic hub.
- collection: [Analysis](https://newruntime.com/analysis.json): Open the latest authorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)