1kmost common free searches/mo
2.5kbiggest one-time free pool (Serper)
$5Brave free credits /mo
500Keiro free /mo
100/dayGoogle Custom Search JSON free
TL;DR · best free tiers in 2026
- **Biggest recurring free pool:** Exa and Tavily, 1,000 searches/month each; Firecrawl 1,000 credits/month. Brave lands ~1,000 queries/month via its $5 credit.
- **Best one-time free bank:** Serper, 2,500 free credits — no monthly reset, but enough to validate and launch.
- **Best no-key free option:** Wikipedia API and DuckDuckGo's HTML endpoint — truly unlimited, no signup, but no SERP structure and ToS gray areas for the latter.
- **Best free for agents:** Keiro (500 free requests/month, search + clean text in one call) and Tavily (1,000/month, AI-optimized). Both are built for LLM/agent loops, not just SERP scraping.
- **The free-tier trap:** every "free" plan is a bounded pool, and agent loops burn pools fast — a deep-research run can eat a month of free quota in one session. Budget a paid tier into the plan from day one.
## What "free" means now
Two structural changes defined the 2026 free-search tier:
1. **Bing died.** Microsoft retired the Bing Search APIs in August 2025, cutting off the old free tier that many tutorials leaned on. Search-API development now happens on independent indexes (Brave) and AI-native players (Exa, Tavily, Firecrawl, Perplexity) rather than the Big Two proxies.
2. **Free became a meter, not a sandbox.** The norm shifted from "X requests/day forever" to "a monthly credit pool." Pools are easier for providers to price, and they reset the moment you hit them.
The upshot: "best free web search API" is now a question about which **monthly pool** is largest for your workload, not which API is unlimited.
## The recurring free tiers (August 2026)
| Provider | Free pool | Resets | What it covers |
|---|---|---|---|
| **Exa** | 1,000 searches/mo | Monthly | Semantic search; contents cost credits |
| **Tavily** | 1,000 searches/mo | Monthly | AI-optimized results, raw-content option |
| **Firecrawl** | 1,000 credits/mo | Monthly | Search + scrape; credit multipliers apply |
| **Brave** | $5/mo in credits (~1,000 queries) | Monthly | Independent index, SERP + Answers |
| **Keiro** | 500 requests/mo | Monthly | Search + clean text (`/search/content`), 3 credits/call |
| **Serper** | 2,500 credits | One-time | Raw Google SERP |
| **SerpAPI** | 250/mo | Monthly | 40+ engines, enterprise SERP |
| **ScrapingDog** | 1,000 credits | One-time | Google SERP + scraping |
| **LLMLayer** | $2 in credits | One-time | Search, scrape, crawl, answers |
| **Jina Reader** | free URL-to-markdown | — | Not search, but the reader half of an agent |
### Exa — 1,000 searches/month
Exa's free tier is 1,000 neural searches per month, which is the biggest recurring pool for semantic search. The catch: search and `contents` (page text) are separate credit pools, so a research agent that searches *and reads* burns both. 1,000 searches/month is plenty to prototype a RAG agent; it is about a week of real research traffic.
### Tavily — 1,000 searches/month
Tavily gives 1,000 searches/month on its free tier, tuned for LLM consumption — source-first results, `include_raw_content` for full pages, and `/extract` for structured data. For agents, this is one of the best free starts in the category because the response is already shaped for a context window rather than for a browser. Paid PAYG starts around $8/1k, so the free 1,000/month is genuinely a month of prototyping.
### Brave — $5/month in credits
Brave moved to a credit model: new users get **$5/month in credits**, which is roughly 1,000 queries at the $5/1k list price (legacy users on the old free plan keep 2,000 queries/month). Brave's pitch is the independent index — no Google or Bing dependency, no tracking, SOC 2 certified — plus the Answers endpoint for grounded, cited responses. The free pool is real but modest, and the lower tiers cap you at 1 query/second, which throttles any throughput-heavy agent.
### Firecrawl — 1,000 credits/month
Firecrawl's free tier is 1,000 credits/month across search, scrape, and crawl. Credits are not pages: JavaScript rendering, JSON extraction, and the Enhanced mode multiply credit cost (commonly ~9 credits for a JSON+Enhanced page). So 1,000 free credits might be ~1,000 simple scrapes or ~110 enriched pages. Still, it is the largest free pool in the "scrape the page, not just find it" category, and the output is LLM-ready Markdown.
### Keiro — 500 requests/month
Keiro's free tier is 500 requests/month, with `/search` at 1 credit and `/search/content` (search **plus** clean page text in one response) at 3 credits. It is the smallest pool on this board, but it is the most agent-shaped: one call returns both the search results and the content an agent needs to read, so there is no second service to buy. 500 requests is enough to build and test a research agent end to end.
### Serper — 2,500 one-time credits
Serper is the value pick that flips the model: **2,500 free credits, one-time**, not monthly. It does not reset — treat it as a free launch bank rather than a recurring source. At list ($1/1k, down to $0.30/1k at volume) it is the cheapest raw Google SERP API that exists, and 2,500 credits is enough to validate an integration and run real traffic while you set up billing. Raw SERPs only — no content, no embeddings — which is exactly right for a finder agent.
### SerpAPI — 250/month
SerpAPI's free tier is 250 searches/month across 40+ engines — Google, Bing, YouTube, Maps, shopping, news. It is the smallest recurring pool here and the API is priced at the premium end ($75+/5k), so the free tier reads as a trial rather than a plan. Choose it for multi-engine coverage and enterprise reliability, not for the free pool.
## The no-key and self-hosted options
If "free" means *actually zero cost and zero signup*, three paths exist:
- **Wikipedia API** — free, no key, genuinely unlimited, and structurally well-suited to agents that answer knowledge questions. The trade: it's one encyclopedia, not the web. The MediaWiki API is a legitimate part of many research stacks.
- **DuckDuckGo HTML** (`html.duckduckgo.com/html/?q=`) — free and keyless, returns HTML results you can parse. ToS and rate-limit risk make it unsuitable for production, but it is the classic zero-budget prototype hack.
- **Google Programmable Search / Custom Search JSON API** — **100 queries/day free**, then $5 per 1,000. A real Google index behind a key, though the free quota is small and results are a subset of Google's index.
- **SearXNG (self-hosted)** — a metasearch engine that aggregates DuckDuckGo, Brave, Google, and others. Costs you a small VPS (~$5–20/mo) and the maintenance, but no per-query bill and no third-party key. The ceiling: the upstream engines can rate-limit or block you, and you are responsible for CAPTCHAs.
## The free-tier trap for agents
Every number above is a bounded pool. Agents are pool incinerators. A single deep-research run — 8 searches, 6 page reads, 4 reasoning hops — can consume 20+ credits in minutes. That is 2% of Exa's monthly pool, or 4% of Keiro's, or the entire daily budget of a constrained setup. If your agent is going to run unsupervised, a free tier is not a plan; it is a demo.
The pattern that works:
- **Prototype** on the free pools — stack Exa (1,000) + Tavily (1,000) + Firecrawl (1,000 credits) + Keiro (500) + Serper (2,500 one-time) and you have a month of agent development for $0.
- **Launch** on a paid tier that matches your request shape. For agents, that means bundling search + content so you don't pay two vendors per hop.
## The bottom line
The best free web search API depends on your workload:
- **Semantic/RAG prototyping:** Exa (1,000/mo).
- **AI-optimized agent search:** Tavily (1,000/mo).
- **Raw Google SERP on a budget:** Serper (2,500 one-time, then $0.30–$1/1k).
- **Search + content in one call:** Keiro (500/mo free).
- **Independent index + privacy:** Brave (~1,000 queries/mo via $5 credits).
- **Full page extraction:** Firecrawl (1,000 credits/mo).
- **Zero key, zero signup:** Wikipedia API; DuckDuckGo HTML for throwaway prototypes.
## FAQ
### What is the best free web search API in 2026?
For agent and RAG workloads, Exa and Tavily both give 1,000 free searches/month, and Firecrawl 1,000 free credits. For a raw Google SERP, Serper's 2,500 one-time credits are the biggest free bank. For search-plus-content in one call, Keiro's 500 free requests/month. "Best" depends on whether you need semantic search, a SERP feed, or full-page content.
### Does Brave Search API still have a free tier?
Yes, but it changed: new users get $5/month in credits, roughly 1,000 queries at list price, instead of the old 5,000-query plan. Legacy free-plan users retain 2,000 queries/month. Lower tiers also cap at 1 query/second.
### Which web search API has the largest free tier?
By recurring monthly pool, Exa and Tavily (1,000 searches each) and Firecrawl (1,000 credits) tie for the lead. By one-time total, Serper's 2,500 credits is the largest single grant.
### Are there truly free web search APIs with no API key?
Wikipedia's API is free, keyless, and unlimited. DuckDuckGo's HTML endpoint is free and keyless but ToS-gray and unsuitable for production. Google's Custom Search JSON API offers 100 queries/day free. SearXNG self-hosted gives you aggregated search for the cost of a small VPS and the maintenance.
### Why did the free tiers shrink so much in 2026?
Microsoft retired the Bing Search APIs in August 2025, pushing developers onto independent and AI-native providers, which then moved from unlimited daily quotas to bounded monthly credit pools. The pattern is now universal: free means "a monthly prototyping allowance," not "unlimited."
### Can I run an AI agent on free search APIs?
Yes, for development: stack Exa + Tavily + Firecrawl + Keiro + Serper and you have a month of free agent building. But agent loops burn quota fast — a deep-research session can consume 20+ credits in minutes — so a production agent needs a paid tier sized to its request shape from day one.