For architects
And the more uncomfortable half of the argument: owning an index used to be the whole pitch, and in 2026 it is table stakes.
Microsoft blocked new Bing Search API resources in early 2025 and retired the service on 11 August, taking Web, Image, News, Video, Entity, Autosuggest, Spell Check and Custom Search with it. Every product built on it had to be rewritten, however good the results had been.
Brave retired its long-standing free developer tier in February 2026, replacing it with a $5 monthly credit. That is a legitimate business decision — and if your onboarding funnel depended on it, it was also your problem overnight.
A reseller's throughput is whatever its upstream allows, shared across every one of its customers. When you scale, you are negotiating for a slice of a quota that belongs to someone else entirely.
In 2024, "we run our own index" was a complete argument. It is not one now. Exa raised $250 million in May 2026 at a $2.2 billion valuation; Parallel Web Systems reached $2.0 billion in April. Both crawl and index the web themselves. Brave has been doing it across 40 billion pages for years.
So owning an index is now the entry requirement rather than the edge. If a vendor's pitch stops there, it is a 2024 pitch. Four things still separate providers:
The own-index providers charge $1 to $7 per 1,000. We charge $0.40, and publish the cost structure that allows it.
Pricing →Every provider ranks documents. None of them let an agent follow the relationships between documents across the open web. We do — six traversal operations.
Coverage graph →Ask any other index why a URL is absent and you get an empty result set. Ask ours and you get a reason.
Coverage transparency →Every own-index provider serves from one jurisdiction on its own terms. Ours deploys inside yours, and at query time nothing leaves.
Sovereign edition →There is one buyer for whom the 2024 argument never stopped being true. A government, a regulator, a public broadcaster or a defence programme building agents on a sovereign model has spent heavily to keep the model at home — and then grounds it through a search API that sends every query abroad. The model is sovereign. The question it asks is not.
Exa, Parallel and Brave own their indexes, and each serves from one jurisdiction on its own terms; none of them can be stood up inside a customer's boundary. Because unlob builds and serves its own index, the whole evidence layer can be. That is what we mean by deployable ownership: the index, the coverage graph, the receipts and the record of what was asked all stay inside your jurisdiction, beneath whichever model you run. It is the one respect in which owning an index is still the edge rather than the entry requirement, and it is the whole of the sovereign edition.
| Provider | Own index? | $ / 1,000 searches | Source |
|---|---|---|---|
Priced: search queries on Gemini 3.x. $14 per 1,000 search queries on Gemini 3.x models after 5,000 free search requests a month (paid tier, shared across Gemini 3.x), plus model token charges. One prompt can trigger several billed queries. Gemini 2.5 models keep the older scheme: $35 per 1,000 grounded prompts after 1,500 free a day. | Yes Google's index | $14 | Gemini API pricing |
Priced: Search plan. Search is $5 per 1,000 requests. The Answers plan is $4 per 1,000 plus $5 per million tokens. | Yes 40B+ pages, no Google or Bing dependency | $5 | Brave Search API pricing |
Priced: paid queries (existing customers only). $5 per 1,000 queries after 100 free a day, up to 10,000 queries a day — for existing customers only. The API is closed to new customers and is discontinued on 1 January 2027. | Yes Google's index, via Programmable Search Engine | $5 | Custom Search JSON API pricing |
Priced: Instant search. Instant search is $4 per 1,000 requests (up to 10 results); Fast and Auto are $7 per 1,000, and the Deep modes $12 to $15. Results beyond 10 add $1 per 1,000, contents $1 per 1,000 pages per content type, and AI summaries $1 per 1,000 pages — so a realistic RAG call costs more than the headline. | Yes Proprietary AI-native semantic index | $4 | Exa pricing |
Priced: Search API with Fast Search. The Search API is $5 per 1,000 requests, or $1 with Fast Search, with no token charges (up to 20 results, 10 by default). Sonar Chat Completions support ended on 27 September 2026 and requests are moving to the Agent API; Sonar charged $5 to $12 per 1,000 by search context size plus $1/$1 per million tokens, and Sonar Pro $6 to $14 plus $3/$15. | Yes Perplexity index | $1 | Perplexity API pricing |
Priced: Turbo processor. The Turbo and Fast search processors are $1 per 1,000; Basic and Advanced are both $5 per 1,000. The Task API, which is what most agent workloads actually use, runs $5 to $2,400 per 1,000 depending on processor. | Yes Proprietary web index and retrieval infrastructure | $1 | Parallel pricing |
$29 per month for 50,000 credits, and a search is one credit. | Yes Our own crawl and index | $0.58 | This page |
$199 per month for 500,000 credits, and a search is one credit. | Yes Our own crawl and index | $0.40 | This page |
Exa, Parallel, Brave, Mojeek, Diffbot and unlob crawl and index the web themselves. Google Custom Search, Gemini grounding and Perplexity serve a first-party index on their own terms. SerpAPI, Serper and DataForSEO resell Google or Bing results, and Tavily and Linkup add a retrieval layer on top of other providers.
Because the results are only the same until they are not. When Microsoft retired the Bing Web Search API in August 2025, every product built on it had to be rewritten regardless of how good the results had been. A reseller cannot promise you continuity it does not have.
Yes — when you specifically need Google's ranking. SEO tooling, rank tracking and any workload where the product is literally "what does Google show" should use a SERP API. The mistake is using one for agent retrieval, where you want good sources rather than Google's ordering of them.
Not on its own, not any more. In 2024 it was the whole argument. By 2026, Exa and Parallel each run their own index with roughly $2 billion valuations behind them. What still differentiates is what you pay per query, whether the API can traverse relationships rather than just rank documents, and whether it will tell you what it excluded.
Smaller than Google, smaller than the well-funded competitors, and bounded deliberately rather than accidentally — we admit a curated core and extend the tail where queries show demand — on the crawler’s schedule, never at query time. The honest framing is that we optimise for precision of inclusion rather than raw size, and that why_not will tell you exactly what is not there.
Yes. Because we build and serve our own index, the whole evidence layer — index, coverage graph and receipts — can be deployed inside a customer boundary: a national cloud, your VPC, or a disconnected environment. At query time nothing leaves. That is the one respect in which owning an index is still a differentiator: Exa, Parallel and Brave own theirs, in one jurisdiction, on their terms. The sovereign edition page sets out the four deployment shapes.
Start free and see whether the coverage holds up for your workload. If it does not, why_not will tell you exactly where.
Evaluating for a team or at volume? Send a workload brief or write to hello@unlob.com — a person replies.