--- title: "The best free web search APIs in 2026" category: "guides" url: https://keirolabs.cloud/blog/best-free-web-search-apis-2026 --- ## The 60-second version Free tiers in web search didn't die in 2026. They got smaller, meaner, and much better at marketing. What changed is the shape of the giveaway: monthly credit pools replaced day-based quotas, credit cards quietly became a condition of entry, and rate limits turned into the real ceiling long before the credits run out. The numbers that decide whether a free tier is useful: - **Tavily** gives 1,000 credits a month, no card, and hard-stops you at zero ([tavily.com/pricing](https://www.tavily.com/pricing)). - **Exa** hands you $20 at signup plus $10 every month, no payment method ([exa.ai/pricing](https://exa.ai/pricing)). - **Serper** still has the biggest one-time bank: 2,500 free queries, no card, no refill ([serper.dev](https://serper.dev)). - **Brave** killed its free plan in February 2026. The replacement is $5 in monthly credits (about 1,000 queries), a card on file, and an attribution string you must display or the credit goes away ([brave.com/search/api](https://brave.com/search/api/)). And us: Keirolabs gives 1,250 credits a month, no card, and our `/search/lite` endpoint costs 0.1 credits a call. That's 12,500 searches a month, free, forever ([keirolabs.cloud/pricing](https://keirolabs.cloud/pricing)). The same API also wins: 15.2 out of 20, first place, on AIMultiple's 2026 agentic search benchmark, ahead of Brave, Firecrawl, Exa, and Tavily. We're in every table below anyway, with the same scrutiny as everyone else. Read the free-tier table, skip to the rate-limit section if you build agent loops, and never wire production traffic to a free tier. ## The September 2026 free-tier table Everything below was pulled from each provider's own pricing page in September 2026, not from anyone's blog post about someone's blog post. | Provider | What's free | Card required? | Refills? | The catch | |---|---|---|---|---| | **Keirolabs** | 1,250 credits/mo = 12,500 `/search/lite` searches | No | Monthly | 30 req/min on free | | **Tavily** | 1,000 credits/mo (~1,000 searches) | No | Monthly | Requests stop dead at zero | | **Exa** | $20 signup + $10/mo credits | No | $10 monthly | ~1,400 searches/mo at $7/1k | | **Firecrawl** | 1,000 credits/mo (~500 searches) | No | Monthly | Search costs 2 credits per 10 results; 2 concurrent, low rate limits | | **Serper** | 2,500 queries, one-time | No | Never | It's a sample, not a tier | | **Linkup** | 4,000 queries at signup + $5/mo credits | Not stated | $5 monthly | Paid rate starts around €5/1k | | **Parallel** | Up to $80 signup + $5/mo credits, plus a free anonymous MCP server | No | $5 monthly | Anonymous MCP runs fast mode at lower rate limits | | **Brave** | $5/mo in credits (~1,000 queries) | **Yes** | Monthly | Attribution required or you lose the credit | | **Google Custom Search** | 100 queries/day (~3,000/mo) | No (API key only) | Daily | Closed to new customers; shuts down Jan 1, 2027 | Two columns do most of the work. "Card required?" separates a sandbox from a trial: Tavily, Exa, Firecrawl, Serper, Parallel and Keiro can't charge you by accident, Brave can. And "Refills?" separates a free tier from a free lunch: Serper's 2,500 queries sound huge until you realize they're all you will ever get. ## What the free pool actually buys Raw credit numbers lie. Providers price in credits, agents think in searches. Here's each free pool converted to actual searches per month, using each provider's own exchange rate: Tavily at 1 credit per basic search, Firecrawl at 2 credits per 10 search results, Exa at $7 per 1,000 `/search` requests, Brave at $5 per 1,000, Parallel's $5 credits at fast-mode $1 per 1,000, and Keirolabs' own-index `/search/lite` at 0.1 credits per search. Keiro /search/lite 12,500 Parallel up to, fast mode 5,000 Google CSE 100/day, dying 3,000 Serper one-time only 2,500 Exa $10 credits/mo 1,400 Tavily 1,000 Brave card + attribution 1,000 Linkup +$4,000 one-time 1,000 Firecrawl search pricing 500 searches per month, gray = one-time or shutting down Read that chart twice. The biggest recurring pool is roughly 12x the most common one. That's a different product philosophy, not a rounding difference: 1,000 searches covers a weekend demo, 12,500 covers a month of building. And the two gray bars are traps dressed as generosity, one because it never refills, one because the service itself expires on [January 1, 2027](https://developers.google.com/custom-search/v1/overview). ## The biggest free tier is also the benchmark leader A big pool would mean nothing if the product behind it were a consolation prize. It isn't. On quality, the same API that gives away 12,500 searches a month is the one that wins: - **AIMultiple's 2026 agentic search benchmark (scored /20):** Keirolabs 15.2, first, ahead of Brave 14.89, Firecrawl 14.58, Exa 14.39, and Tavily 13.67. - **FinanceBench:** Keiro 78%, first, ahead of Valyu at 73%, Parallel at 67%, Exa at 63%, and Google at 55%. - **Head to head against Parallel's turbo mode:** `/search/lite` won 91 of 100 head-to-head comparisons, took the composite score 341.6 to 170.4, and swept 6 of 9 query archetypes ([full methodology and per-archetype results](https://keirolabs.cloud/bench/keiro-lite-vs-parallel-turbo)). So the usual trade in this category, generous free tier versus serious API, doesn't apply here. The common pattern is a provider picking one: a famous free pool attached to a middling ranker, or a benchmark leader that meters you from request one. Keiro holds both ends at once: the largest recurring no-card pool in the category and the top benchmark scores to spend it on. ## The fine print, provider by provider ### Tavily: 1,000 credits a month, then silence Tavily's free plan is the cleanest deal in the category: 1,000 credits monthly, no card. The fine print is the behavior at zero. "If you exhaust your monthly credits, requests will stop until your credits reset or you upgrade" ([Tavily pricing](https://www.tavily.com/pricing)). That's the honest version of a free tier: a hard wall you can't fall through. After the wall, pay-as-you-go runs $0.008 per credit, $8 per 1,000 basic searches ([Tavily docs](https://docs.tavily.com/documentation/api-credits)), the highest recurring search rate on this list. Tavily knows the free tier is marketing, and prices accordingly. ### Exa: the best drip in the business $20 in credits at signup, $10 every month, no payment method, 10 queries-per-second on free ([Exa pricing](https://exa.ai/pricing)). At Exa's `/search` rate of $7 per 1,000 requests plus $1 per 1,000 results beyond 10 ([Exa pricing docs](https://exa.ai/docs/reference/pricing)), the $10 drip is about 1,400 searches a month. The drip keeps an account alive and testing for months. Watch the search-plus-contents combination though: contents calls add credits on top of the search call. ### Firecrawl: 1,000 credits that shrink on contact Firecrawl gives 1,000 credits monthly with no card ([Firecrawl pricing](https://www.firecrawl.dev/pricing)), and it's the one free tier where the exchange rate punishes the exact thing you signed up for. Scraping costs 1 credit per page, but **search costs 2 credits per 10 results**, so 1,000 credits is about 500 searches, not 1,000. JSON extraction adds 4 credits per page. The free plan also runs 2 concurrent requests at the bottom of their rate-limit ladder, about 2 requests per minute. Firecrawl is a scraping company with a search endpoint; if search is your workload, their credits buy half of what the sticker says. ### Serper: the one-time bank 2,500 free queries, no credit card ([serper.dev](https://serper.dev)). That's the entire free offer: no monthly reset, no drip. Enough to validate an idea or run a benchmark, not enough to run anything for a month. When the bank empties, packs start at $50 for 50,000 queries, $1.00 per 1,000, credits valid 6 months ([serper.dev](https://serper.dev)). Serper's real product is raw Google SERP at the lowest price per query in the industry, and the 2,500 free queries are a tasting menu. ### Linkup: 4,000 queries, then $5 a month Linkup advertises "run 4,000 queries for free" ([linkup.so/pricing](https://www.linkup.so/pricing)), and per [Parallel's September 2026 roundup](https://parallel.ai/articles/best-free-web-search-api) new accounts get 4,000 queries on signup, then topped up to $5 in credits each month. At Linkup's paid rate of around €5 per 1,000 standard searches, the monthly drip is roughly 1,000 queries. The signup bank plus a recurring drip is the most generous structural combo after Keiro's; the gap is that Linkup's pricing page doesn't state card requirements in either direction, so this post flags that as unverified rather than guess. ### Parallel: the only zero-key option Parallel is the most interesting free offer of 2026 because part of it needs no account at all. The Search MCP server at `search.parallel.ai/mcp` is free and usable anonymously, no API key, running fast mode with lower rate limits ([Parallel MCP docs](https://docs.parallel.ai/integrations/mcp/search-mcp)). If you just want web search inside an MCP client, right now, that's the zero-friction answer. On top of that, accounts can earn up to $80 at signup plus $5 in free credits monthly, and startups up to $250 ([Parallel pricing](https://parallel.ai/pricing)). Parallel prices per request, $0.001 to $0.005 for a 10-result request, so the $5 drip is worth 1,000 to 5,000 searches depending on mode. ### Brave: the free tier that became a coupon This is the cautionary tale of the year. Brave's free plan was 2,000 queries a month at launch in 2023, raised to 5,000 by August 2025, hard-capped, with the card collected "only to confirm your identity and will not be charged." In February 2026 the free plan disappeared. The replacement: every plan includes $5 in monthly credits, about 1,000 queries at $5 per 1,000 requests, and the card that was never supposed to charge "becomes an active billing instrument" the moment you exceed them ([Brave Search API](https://brave.com/search/api/), [implicator.ai's account](https://www.implicator.ai/brave-drops-free-search-api-tier-puts-all-developers-on-metered-billing/)). One more string: "to take advantage of this free credit, all you need to do is attribute the Brave Search API in your project's website / about pages." Drop the attribution, drop the credit. Brave still runs a great independent index. But "free" now means a $5 coupon that requires a footer link and a card on file. ### Google Custom Search: the door that closed 100 free queries per day, $5 per 1,000 beyond that ([Google's Custom Search overview](https://developers.google.com/custom-search/v1/overview)). On paper that's ~3,000 free searches a month, the second-best recurring pool here. In practice the API was [announced as discontinued in January 2026](https://www.octoparse.com/blog/google-official-search-api), is closed to new customers, and [existing customers have until January 1, 2027](https://developers.google.com/custom-search/v1/overview) to move off. Do not build anything new on it. ## Paid rates after the free runs out The number that actually matters isn't the free tier. It's the rate you pay on request 1,001, because that's where every free-tier story ends. Here's what the meter charges once the free pool is gone, from each provider's published rates: | Provider | Paid rate after free | Cheapest paid entry | Rate source | |---|---|---|---| | **Keirolabs** | $0.25/1k headline on `/search/lite` (0.1 credit/request on monthly plans: $0.24/1k on Essential, $30/mo = 125,000 lite searches, down to $0.13/1k on Pro and $0.08/1k on Startup) | $10 Starter pack, 1,500 credits, no subscription; packs bill lite at 0.5 credits/request ($2.50 to $3.33/1k) | [keirolabs.cloud/pricing](https://keirolabs.cloud/pricing) | | **Serper** | $1.00/1k (Starter pack), down to $0.30/1k at 12.5M credits | $50 for 50,000 queries | [serper.dev](https://serper.dev) | | **Parallel** | $1.00 to $5.00 per 1k requests, by mode | Pay per request, no plan | [parallel.ai/pricing](https://parallel.ai/pricing) | | **Firecrawl** | ~$1.66/1k to ~$6.40/1k searches by plan (2 credits per 10 results) | $19/mo Hobby, 5,000 credits | [firecrawl.dev/pricing](https://www.firecrawl.dev/pricing) | | **Brave** | $5.00/1k (Search) | Card on file, credits auto-apply | [brave.com/search/api](https://brave.com/search/api/) | | **Linkup** | from ~€5/1k standard searches | Pay per use | [linkup.so/pricing](https://www.linkup.so/pricing) | | **Exa** | $7/1k `/search` + $1/1k extra results | Pay from credit balance | [exa.ai/docs/reference/pricing](https://exa.ai/docs/reference/pricing) | | **Tavily** | $8.00/1k basic searches (pay-as-you-go $0.008/credit) | Pay-as-you-go, no plan | [docs.tavily.com](https://docs.tavily.com/documentation/api-credits) | | **Google CSE** | $5/1k, until the service dies | Not available to new customers | [developers.google.com](https://developers.google.com/custom-search/v1/overview) | Sort that column and a pattern appears: the providers with the most famous free tiers are often the most expensive after them. Tavily's free 1,000 is famous. Tavily's $8/1k pay-as-you-go is the highest recurring search rate on this list. Free tiers are customer-acquisition budgets, and the acquisition cost gets recovered on the other side. Pick a provider on the paid rate; treat the free tier as a runway calculation. One note on how the Keirolabs meter works, because the two paths get conflated. The $0.25 per 1,000 headline is the plan rate: on monthly plans, `/search/lite` bills at 0.1 credits a request, which works out to $0.24 per 1,000 on Essential, $0.13 on Pro, and $0.08 on Startup. If you don't want a subscription at all, one-time credit packs are the pay-as-you-go path: Starter $10 for 1,500 credits, Growth $30 for 5,000, Scale $100 for 20,000, all valid 6 months. Packs bill lite at 0.5 credits a request, which works out to $2.50 to $3.33 per 1,000 searches, still below every per-1k rate in the table except Keiro's own plan rates. The rest of the catalog prices in whole credits: `/search/fast` at 1, `/search/content` and `/extract` at 3, `/answer` at 5, `/agentic` at 20, and `/batch` at 1 credit per query. ## Rate limits bind before credits do Here's what almost every free-tier comparison gets wrong: for agent workloads, the credit pool is usually not the binding constraint. The rate limit is. Work the numbers. Firecrawl's free plan runs 2 concurrent requests at about 2 requests per minute ([Firecrawl pricing](https://www.firecrawl.dev/pricing)). Even if the credits were infinite, a deep-research agent that fires 30 calls a minute will spend its afternoon in backoff. Exa's free tier allows 10 QPS, genuinely generous for bursty loops ([exa.ai/pricing](https://exa.ai/pricing)). The Keiro free tier runs 30 req/min: an agent can burn all 12,500 monthly `/search/lite` searches in about 7 hours of sustained maximum, the right order of magnitude for a month of development. Google's 100-a-day cap was the purest example: it was never a credit limit, it was a rate limit denominated in days, which made its "3,000 free searches" behave like 3,000 spread across 30 calendar days, useless for a launch-day spike. The rule the Keirolabs team uses internally: budget your free tier in requests-per-minute times minutes-of-work, not in monthly credits. If your agent needs 40 concurrent calls and your free plan allows 2, the 1,000-credit number on the pricing page is decoration. > Every free tier is priced to cover the demo, not the deployment. The moment your usage pattern stops looking like a demo, the rate limit, not the credit pool, is what stops you. ## A card on file changes the risk The second structural shift of 2026 is quieter: cards moving from optional to required, and from inert to armed. Brave has required a card since 2023, and for three years the FAQ said the card "will not be charged" on free plans ([implicator.ai](https://www.implicator.ai/brave-drops-free-search-api-tier-puts-all-developers-on-metered-billing/)). In February that same card became a live billing instrument with no published spending cap on the pricing page. Nothing about Brave's signup flow changed on the day the semantics flipped. That's the whole risk in one story: a hard-capped free tier is a contract you can read, a card-backed credit balance is a contract the provider can rewrite. Firecrawl runs the same mechanic one tier up: on paid plans, when your credits run out the system automatically adds them in $5 increments charged to the card on file, up to a limit you set ([Firecrawl pricing](https://www.firecrawl.dev/pricing)). Auto-topup is a reasonable product. But it means "free tier" plus a stored card plus a forgotten agent loop equals a bill you find out about from your statement. We don't take a card for the free tier. 1,250 credits, 30 req/min, no payment method, and the pool refills every month ([keirolabs.cloud/pricing](https://keirolabs.cloud/pricing)). We'd rather have 10,000 developers poking the API than 500 watching a meter. Judge us the same way you judge everyone else in this post, by what comes after the free credits: on monthly plans, `/search/lite` runs $0.25 per 1,000 searches at the headline rate, $0.24 per 1,000 on Essential and $0.08 per 1,000 on Startup, and the no-subscription packs bill the same endpoint at $2.50 to $3.33 per 1,000. The same API that tops AIMultiple's agentic benchmark at 15.2 out of 20 and took 91 of 100 head-to-heads against Parallel turbo is the one handing out the 12,500 searches. Our practical rule: free tier with no card is a sandbox, play freely. Free credits with a card on file is a trial with a tripwire, set your own external alarm before your agent does. ## How to stack three free tiers into a real month You can assemble a genuinely useful zero-budget month by refusing to marry one provider. A stack Keirolabs would actually run: - **Default search: Keiro free, 12,500 `/search/lite` searches.** No card, 30 req/min. This is the workhorse pool; at 400 searches a day it survives 31 days. - **Deep reads: Tavily free, 1,000 credits** for its extract-style content pulls, or **Firecrawl free, 1,000 credits** when you need raw page scraping and the 2 req/min pace is tolerable. - **Zero-setup fallback: Parallel's anonymous Search MCP.** No key, no account; when your main key dies mid-demo, this keeps the demo alive. That's a real month of development: 13,500+ searches plus a scrape pool, at $0, no card anywhere. What it is not: production. Free tiers have no SLA, rate limits that change without ceremony, and Brave demonstrated this year that the terms can be rewritten in a single blog post. The mature move is to prototype on the stack, then graduate your mainline traffic to a paid rate you've already priced. The best free tier is the one you can leave in one afternoon. ## The boring takeaway "Free" in 2026 means three separate questions: how big is the pool, does a card sit behind it, and how fast can you draw from it. Recurring beats one-time for anything longer than a week. No-card beats card-on-file unless you like reading your own bills as a surprise. And rate limits bind before credits do, so price your workload in requests per minute before you get attached to a credit number. Keirolabs built its free tier the way it wished others had: big pool, no card, hard caps instead of metered billing, 12,500 lite searches a month. Grab a key at [platform.keirolabs.cloud](https://platform.keirolabs.cloud), run your workload against it for a month, and if you outgrow it, the paid rates are published in plain numbers on [the Keirolabs pricing page](https://keirolabs.cloud/pricing). That's the whole pitch.