Search APIs split: commodity calls deflate, 'research effort' inflates
Agent-facing search APIs are splitting into two price layers. The commodity call has converged at $5–7 per 1,000 requests — You.com, Exa, Linkup and Groq all land in the band — and is deflating, while a research layer above it prices by compute effort at 10–400× per call: Linkup's 10× deep search, Tavily's 250-credit pro research requests, You.com's $12/1k-to-$2,000+/1k Research ladder.
What's happening — and why
What's happening: four search-API vendors now run an explicit two-layer price card. The bottom layer is the commodity call — You.com Search at $5/1k, Exa at $7/1k, Linkup standard at $5/1k, Groq's built-in web search at $5–8/1k — and it keeps getting cheaper: in April 2026 You.com cut its Contents API 10×, from $10 to $1 per 1,000 pages. The top layer prices effort. You.com's Research ladder runs $12/1k up to a contact-sales Frontier tier above $2,000/1k; Exa Agent bills on auto effort or four fixed-effort modes from $0.025 to $2.00 per request (an 80× spread); a Tavily research call with model=pro can burn up to 250 credits (~$2.00) against a 1-credit basic search; Linkup's deep search costs exactly 10× standard.
Why: the two calls have nothing in common under the hood. A commodity search returns indexed results at near-zero marginal cost, so it competes like a utility and converges on a band — Groq, an inference platform, independently landed on the same $5–8/1k range. A research call spends real reasoning compute: the model plans, searches repeatedly, reads and synthesizes, so the cost driver is the effort budget, not the data returned. Pricing the layers separately lets vendors race to the bottom on retrieval while rebuilding margin on research — You.com cut its commodity prices and raised Research Lite to $12/1k within the same quarter. Vendors with no reasoning compute to sell (SerpApi, Oxylabs) stay on pure volume ladders and compete on deflation alone.
How it works
Evidence over time
15 supporting · 3 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| You.com | Apr 2026 | Cut commodity API prices sharply — Contents from $10 to $1 per 1k pages (10x), Search to a flat $5/1k. |
| You.com | Jun 2026 | Raised Research Lite to $12/1k and launched a dedicated Finance Research API; the Research endpoint ladder runs $12/1k up to a >$2,000/1k contact-sales Frontier tier — effort, not data volume, sets the price. |
| Exa | Apr 2026 | Restructured pricing into per-endpoint cards and raised base Search to $7/1k; Exa Agent bills on auto effort or four fixed-effort modes from $0.025 to $2.00 per request. |
| Tavily | Jun 2026 | Basic search costs 1 credit; a Research request with model=pro can burn up to 250 credits (~$2.00 PAYG) — a >100x in-product effort ladder. Failed extractions cost 0 credits. |
| Linkup | Jun 2026 | Deep search costs 10x a standard search ($0.05 vs $0.005); the per-request unit price of standard search has never moved — only packaging changed. |
| Groq | Jan 2026 | Priced built-in web search for agents at $5–$8 per 1,000 requests — an inference platform independently landing on the same commodity search price band. |
| Exa | Jul 2026 | Recut the effort ladder in both directions and gave away the commodity layer: free tier raised from 1,000 to 20,000 requests/month (20×); Agent repriced from $0.025–$2.00 to $0.012–$1.00 per run, halving the ceiling and adding a new Minimal fixed-effort mode at $0.012; Agent Compute Units at $0.10/ACU and in-agent search cut from $0.007 to $0.005. Core endpoints unchanged — Search $7/1k, Deep $12–15/1k, Contents $1/1k pages, Monitors $15/1k, Answer $5/1k. |
| You.com | Jul 2026 | Put a firm price on the top of the effort ladder and cut it ~40%: the Research API's Frontier tier changed from '>$2,000.00 /1k calls' to '$1,200.00 /1k calls', keeping the Contact Sales CTA. Against You.com's own $5/1k Web Search rate that is a published 240× spread on one price list — the effort ceiling is now a number a buyer can model rather than an open-ended label. |
| Linkup | Jul 2026 | Added a new effort UNIT — per task, not per call: a closed-beta Extract endpoint (structured crawl) at a variable ~$2–$10 per task with a $10 minimum balance, two to three orders of magnitude above its $0.005 standard search. Simultaneously corrected the marketing page's Deep-search range from $0.05–$0.55 to $0.05–$0.055, matching the billing docs — a published 10× overstatement of its own effort ceiling. All other per-request rates unchanged. |
| Exa | Jul 2026 | The commodity layer stops being metered at all: Exa dropped the '20,000 requests/month' framing from the pricing page and docs in favour of a pure dollar-credit model — signup credit doubled $10 → $20, and the recurring monthly free credit rose $7 → $10 and is now granted to EVERY Free Tier account with no payment method required (previously card-gated). Monthly credits expire at month end. |
| You.com | Jul 2026 | Commodity search goes free at the daily level: the Web Search API pricing card now shows 'Free / 100 calls per day', with the $5.00 per 1,000 calls rate applying only after that daily quota — replacing a flat meter that had no daily allowance beyond the general $100 account credit. (Docs had not caught up at capture.) |
| Firecrawl | Jul 2026 | Gives the commodity call away uniformly: every plan card — Free, Hobby, Standard, Growth, Scale and Enterprise — now lists a flat '+1,000 search credits' line beneath its main credit allotment, identical regardless of plan size, on top of the unchanged 2-credits-per-10-results Search rate. Previously Search consumed the single per-tier pool (e.g. Hobby's 5,000 credits/month). |
| Perplexity AI | Aug 2026 | The commodity floor broke below the $5-7/1k band, and the cut was selective. Perplexity's Agent API halved BOTH web_search (to $0.0025 per invocation, i.e. $2.50/1,000) and fetch_url (to $0.00025) on 2026-07-29. As of the 2026-08-14 capture, web_search is still $0.0025 while fetch_url has reverted to its original $0.0005 — confirmed independently of the rate table by the docs' own worked 'Agent API Research Preset' example, whose total moved from $0.00675 to $0.007, a difference that reconciles only if fetch_url doubled back. people_search and finance_search hold at $0.005 and the $0.03 per-session sandbox fee is unchanged. A vendor that reverses one tool price and keeps the other is repricing a workload, not running a promotion: search got permanently cheaper, retrieval did not. |
| You.com | Aug 2026 | Synthesis priced as a commodity by the vendor with the steepest effort ladder in the corpus. You.com launched an Answer API as a fifth priced endpoint — one call that runs a web search and returns a synthesized, citation-grounded answer generated only from retrieved passages, using the same query operators as Web Search and claiming 93.48% accuracy on SimpleQA — at $5.00 per 1,000 calls, IDENTICAL to the raw-results Search API rate. You.com already prices Research from $12/1k (Lite) to $1,200/1k (Frontier) and Finance Research from $110 to $500/1k; it declined to put its newest reasoning product anywhere on that ladder. All other rates unchanged. The layer boundary moved down: reasoning alone no longer earns the effort premium, only multi-step research does. |
| Diffbot | Aug 2026 | A commodity-only entrant charging nothing for search. Diffbot added Web Search as a fifth credit-metered product at 1 credit per query — the same rate as a full page extraction — folded into the existing 'all APIs included' entitlement on every tier including Free, with Free/Startup/Plus/Enterprise unchanged at $0/$299/$899/custom and Crawl still the sole feature gated to $899+. The relaunch also exposed a previously undocumented mechanic: the Free plan does not auto-bill overage — exceeding 10,000 credits/month returns a 429 until reset — while paid plans bill overage pro rata. A vendor with no effort tier to protect adds search and prices it at zero marginal cost, matching Firecrawl's +1,000-credits-on-every-plan move six days earlier. |
Counterexamples
- SerpApi · Jun 2026 — No effort dimension at all: a pure volume ladder of successful searches running to 'Cloud 54M' at $106,050/month — the classic per-call model, priced by quantity, not compute.
- Oxylabs · Jun 2026 — Commodity-side deflation without a research tier: running 40% off Web Unblocker for six months — scraping infrastructure competes on unit price alone.
- Oxylabs · Aug 2026 — The counterexample held byte-for-byte through two full page restructures in eight days. Oxylabs restored its Web Scraper API's two-tab Regular/Enterprise layout on 2026-08-04 (bringing back Free Trial, Advanced $249/mo, Venture $499/mo and Corporate $2,000/mo) and then flattened it back to a single four-column grid on 2026-08-11 (Micro $49, Starter $99, Business $999, Custom+ from $0.25/1K). Every per-1K rate — including the Amazon/Google/other target splits, the JavaScript-rendering surcharge, the media-download fee and the top-up ceilings — is identical across the flattened, restored and re-flattened captures. A vendor moving bytes has only a volume ladder to rearrange.
Trivia
-
You.com moved its prices in both directions inside one quarter: in April 2026 it cut the Contents API 10x ($10 → $1 per 1,000 pages) and Search to a flat $5/1k — then on 2026-06-01 raised Research Lite to $12/1k and shipped a dedicated Finance Research API. Commodity down, effort up, same vendor, eight weeks apart.
-
In the same month (April 2026) that You.com cut Search to $5/1k, rival Exa raised its base Search to $7 per 1,000 — direct competitors crossing in opposite directions while both built out effort-priced tiers above the base call (Exa Agent's fixed-effort modes span $0.025 to $2.00 per request, an 80x ladder).
-
The single most expensive call on Tavily's price list (verified 2026-06-03) is a Research request with model=pro, which can burn up to 250 credits (~$2.00) in one request — more than 100x its basic 1-credit search. You.com's Research API stretches further still: $12/1k (lite) to a contact-sales "Frontier" tier listed at more than $2,000/1k — a ~167x spread on a single endpoint.
-
You.com put a firm number on its most expensive endpoint and it came down 40%. The Frontier research tier's label changed from ">$2,000.00 /1k calls" to "$1,200.00 /1k calls" on 2026-07-22, with the Contact Sales CTA retained. Against its own $5/1k Web Search rate that is a published 240x spread on one vendor's price list — and it is now a price rather than a threat.
-
The commodity call went free at three vendors in three weeks. You.com added 100 free Web Search calls PER DAY before the $5/1k rate applies (2026-07-29); Exa dropped its request cap entirely for pure credits — $20 at signup, $10 every calendar month, no payment method required (2026-07-29, up from $10 and $7 with a card on file); Firecrawl bundled an identical +1,000 search credits into every plan from Free to Enterprise (2026-07-30). Deflation reached zero.
-
Exa halved its effort ceiling while adding a cheaper bottom rung: Agent moved from $0.025–$2.00 to $0.012–$1.00 per run on 2026-07-14, with a new Minimal mode at $0.012. Five fixed-effort modes now span 83×, and auto effort meters Agent Compute Units at $0.10/ACU plus $0.005 per in-agent search. The ladder is getting longer at the bottom and shorter at the top.
-
Linkup published a 10× overstatement of its own effort ceiling and then corrected it. Its marketing page advertised Deep search at $0.05–$0.55 while the billing docs capped it at $0.05–$0.055; the 2026-07-14 change aligned both on $0.05–$0.055. In the same release it introduced a closed-beta Extract endpoint at a variable ~$2–$10 per TASK with a $10 minimum balance — a new unit two to three orders of magnitude above its $0.005 standard search.
For buyers
Price agent workloads by separating retrieval from research before talking to any vendor. For raw results, $5–7 per 1,000 calls is the benchmark — and it's falling, so treat the band as a ceiling in negotiation, not a floor. For research endpoints, the meter is the vendor's compute budget, not your data volume: one Tavily model=pro request can burn 250 credits (~$2.00) — what 250 basic searches cost — and You.com's Frontier research tier is priced around 400× its own $5/1k commodity call. Cap effort explicitly in agent code: pin Exa Agent to a fixed-effort mode rather than auto, set credit limits on Tavily, and decide per task whether Linkup's 10× deep search earns its premium. Note what doesn't bill, too — Tavily charges 0 credits for failed extractions; ask every vendor the same question.
For vendors
Selling the effort layer requires metering it: per-endpoint rate cards (Exa restructured into exactly that in April 2026), named effort modes with fixed prices ($0.025–$2.00 at Exa), or a credit burn that scales with the work (1 to 250 credits at Tavily), plus a contact-sales tier for unbounded jobs (You.com's $2,000+/1k Frontier). The commodity call becomes the acquisition surface — keep it inside the $5–7/1k band and let it deflate. The honest test is whether you run meaningful reasoning compute per call: if not, you're SerpApi, and the only ladder you can sell is volume — its published tiers run to $106,050/month — with deflation (Oxylabs' six-month 40%-off on Web Unblocker) as the competitive weapon.
Outlook — what to watch
First logged June 2026 at a corpus of 207 (now 338), off six evidence points across five vendors. Expect the spread to widen: commodity search has You.com's 10× Contents cut as precedent for further deflation, while research ladders keep adding rungs — You.com raised Research Lite to $12/1k and shipped a dedicated Finance Research API in June 2026, making vertical-specific research endpoints the likely next rung. The trend sharpens if more vendors publish named effort modes or the commodity band breaks below $5/1k; it weakens if research pricing collapses back toward flat per-call rates, or if buyers reject unpredictable effort bills and force fixed-price research SKUs. Watch SerpApi: if a pure volume player adds an effort-priced endpoint, the bifurcation is the whole market.
Bottom line
Four search-API vendors — You.com, Exa, Tavily, Linkup — now run a two-layer price card: a commodity call converging at $5–7 per 1,000 requests and deflating, and a research layer priced by compute effort at 10–400× per call. Benchmark retrieval against the band, and cap effort modes in agent code before one deep-research call costs you 400 searches.
FAQ
How much does a web search API cost in 2026?
The commodity per-call layer has converged at $5–7 per 1,000 requests: You.com Search at a flat $5/1k, Exa at $7/1k, Linkup standard at $5/1k ($0.005 per request), and Groq's built-in agent web search at $5–8/1k. The band is deflating — You.com cut its Contents API from $10 to $1 per 1,000 pages in April 2026.
What is effort-based pricing for search and research APIs?
Pricing where the meter is the compute the vendor spends on your task, not the number of calls or the data returned. Exa Agent offers four fixed-effort modes from $0.025 to $2.00 per request, a Tavily research call with model=pro can burn up to 250 credits (~$2.00) versus 1 credit for a basic search, and You.com's Research endpoints ladder from $12/1k to a contact-sales Frontier tier above $2,000/1k.
Why do deep-research API calls cost 10–400× more than a search?
Because a research call runs reasoning compute — the model plans, searches repeatedly, reads and synthesizes — while a commodity search just returns indexed results. Linkup prices the gap at exactly 10× ($0.05 deep vs $0.005 standard); Tavily's model=pro research runs over 100× a basic search; You.com's $2,000+/1k Frontier tier sits around 400× its $5/1k commodity call.
Do all search API vendors price by effort?
No. Effort pricing appears only where the vendor runs meaningful reasoning compute per call. SerpApi sells a pure volume ladder of successful searches running to a published 'Cloud 54M' tier at $106,050/month, and Oxylabs competes on unit-price deflation alone — it ran 40% off Web Unblocker for six months. Vendors that can't sell effort stay on volume.