TL;DR
- I ran a head-to-head benchmark across four search APIs built for agents — Exa, Perplexity Search API, Parallel, and Amazon Bedrock AgentCore. There's no single winner: Exa leads on primary-source coverage, Perplexity on useful-result density, AgentCore on latency.
- Tavily, a search-for-agents startup that had raised $25M total, got acquired by Nebius for $275M in February 2026 — a ~98x revenue multiple most "which search API is best" comparisons never mention.
- No analyst firm tracks "search API for agents" as its own category yet. The closest proxy, Fact.MR's agentic tool-use/API-integration estimate, puts the whole adjacent category at $6.9B in 2026 — and that's every agent-tool-integration dollar, not search specifically.
- "Search API for agents" is actually four different businesses wearing the same label: agent-native search, raw SERP scrapers, platform-bundled grounding, and enterprise/consumer search that isn't open-web at all.
- Exa, Perplexity, and Parallel have all quietly built a second, much more expensive pricing tier for agent/task products on top of their raw search API — the market is redefining "search" as the cheap fraction of a bigger, per-task unit.
I benchmarked four search APIs built for AI agents — Exa, Perplexity's Search API, Parallel, and Amazon Bedrock AgentCore Web Search — across 30 research queries, live filters, page-fetch behavior, and a fixed-model review of 600 saved results. No provider won outright:
| Provider | Useful results (/150) | Primary-source results (/150) | Full-facet coverage (/90) | Median latency | Est. cost, 30 searches |
|---|---|---|---|---|---|
| Exa | 140 | 92 | 68 | 1.807s | $0.21 |
| Perplexity Search | 145 | 69 | 66 | 1.750s | $0.15 |
| AgentCore Web Search | 124 | 76 | 58 | 0.883s | $0.21 |
| Parallel Basic | 112 | 60 | 44 | 0.979s | $0.15 |
Exa returned the most primary-source evidence. Perplexity returned the most useful results at the lowest search cost. AgentCore was fastest by a wide margin — because it runs its own index instead of querying someone else's.
That result is the boring part. The interesting part showed up once I looked past the benchmark and into what's happening around these four vendors: funding rounds sized like infrastructure companies, a completed acquisition, and pricing tiers that don't map to the "$X per 1,000 queries" number any of them lead with.
The money says this is a real market now, not a feature
Line up the last two years of funding across the vendors in this space and the pattern is infrastructure-company money, not feature money:
| Company | Latest disclosed raise | Valuation | Date |
|---|---|---|---|
| Perplexity | in talks for new round | >$30B (proposed) | Aug 2026 |
| Exa | $250M Series C | $2.2B | May 2026 |
| Parallel | $100M Series B | $2B | Apr 2026 |
| Tavily | acquired, not raised | $275M acquisition | Feb 2026 |
Exa's Series C came eight months after an $85M Series B. Parallel — founded by Parag Agrawal — raised two $100M rounds five months apart. Perplexity's valuation ladder went from $520M in January 2024 to $22.6B in January 2026 before the reported Nvidia talks, on the back of annualized revenue that tripled from under $250M to over $750M in about seven months.
The clearest signal is Tavily. It was a two-year-old company that had raised only $25M total when Nebius acquired it for $275M in February 2026 — roughly a 98x multiple on 2025 revenue. Tavily's own founder framed the deal in exactly these terms: "agentic search is a multi-billion-dollar opportunity, and we believe the market is poised to grow exponentially as enterprises deploy autonomous AI systems." An AI-infrastructure company didn't buy a feature. It bought a search layer to bundle into its own cloud platform.
None of this shows up if you only compare list prices. It shows up when you ask who's writing the checks.
Nobody has drawn the category boundary yet
Here's what surprised me: there's no dollar figure for "search API for agents" as its own line item, anywhere. The closest analyst estimate, Fact.MR's agentic AI tool-use and API-integration market, sizes the whole category — every agent tool call, not search specifically — at $6.9B in 2026, growing to $39.6B by 2036. Two different firms publish an "AI Search Engine market" estimate under the same name and disagree by more than 3x on the ten-year number, and neither one is measuring the developer-facing infrastructure layer this post is about — both are sized around consumer and enterprise AI search products.
That gap matters more than it looks. When a category has no agreed boundary, every vendor gets to draw its own — which is exactly what's happening in the next section.
It's four different businesses, not one
Line up Exa, Tavily, Perplexity Search, Parallel, SerpAPI, Bing Grounding, and Google's Search grounding side by side and they stop looking like competitors. They're four different products that all get called "search API":
| Shape | Examples | Billing unit | What you get |
|---|---|---|---|
| Agent-native search | Exa, Perplexity Search, Parallel, Tavily | Per developer-issued call | Cleaned, ranked context sized for a prompt window |
| Raw SERP scraper | SerpAPI | Per developer-issued call | Unprocessed results-page JSON |
| Platform-bundled grounding | Bing Grounding, Google Search grounding | Per model-triggered query | Grounding as a feature inside a bigger AI platform |
| Adjacent, not comparable | Glean | Per seat | Answers from a private internal index, not the open web |
Tavily's own docs describe the first row precisely: "the web layer for AI agents." It also has a named partnership as LangChain's recommended search tool — the strongest adoption signal I found for any single vendor. The second row is easy to mistake for the first until you read a direct comparison: "A SERP API returns the raw results-page JSON... An agent search API... returns a short list of already-cleaned, ranked context... They solve different problems and aren't interchangeable." SerpAPI isn't competing with Exa. It's a different tool that happens to share a category label. And Glean has nothing to do with the open web at all — it sits in the same sentence as Exa in a lot of vendor comparisons and shouldn't.
If you're picking a vendor by scanning a feature matrix, you're comparing across all four rows at once without realizing it.
The pricing model is splitting in two
The clearest evidence that this market is maturing is what's happening to price. Three of the four vendors I tested now sell raw search at one rate and an "agent" or "task" tier at a rate that isn't a small multiple — it's a different order of magnitude.
Parallel's Search API runs $1–5 per 1,000 calls. Its Task API runs $5 to $2,400 per 1,000 completed task runs, priced per finished run regardless of how much it did to get there. Perplexity's Agent API decouples further — flat per-tool-call fees on top of direct model token rates, with no markup on the model itself. Exa's own launch post for its Deep search endpoint states the thesis outright: "search and agents are converging... the best way to answer a complex query isn't a single search, it's an agent."
"Cost per completed task" is becoming the standard unit for this, not "cost per query." Artificial Analysis's Search API Index already reports cost per 1,000 benchmark tasks, split into search and model spend, as a named metric across its whole suite — not something invented for one vendor.
What's missing
My own benchmark is priced entirely in the old unit. Every number in it — $7 per 1,000 queries for AgentCore, $5 for Perplexity Search, $7 for Exa — measures cost per query. The vendors' own agent-tier products are priced per completed task, and I haven't run a controlled comparison in that unit. So the two most concrete numbers in this post — my own head-to-head benchmark and the market's newest pricing convention — are measuring two different things, and I don't yet have a bridge between them.
So what
If you're choosing a search API for an agent right now, the vendor comparison you actually need isn't "who scores highest." It's three separate questions: which of the four product shapes does this vendor actually sell, is the number on the pricing page a per-query or per-task rate, and does the pricing model match how many calls your agent will actually make. Get any one of those wrong and you're comparing a SERP scraper to an agent-native API, or a per-query estimate to a per-task bill.
The money already moved. Tavily sold for 98x revenue and Exa, Perplexity, and Parallel are all raising at multi-billion-dollar valuations on the same premise Tavily's founder stated directly — agents will make more search calls than any human ever did, and whoever owns that layer owns real infrastructure. The benchmark numbers tell you who's fastest today. The funding and pricing data tell you this stopped being a feature comparison months ago.
Part 1 of the Search for Agents series. Part 2: What actually differs, feature by feature →
Top comments (0)