Skip to main content
Private access is opening up — request an invite
askFinz
Compare

Most AI productsrenttheir search results.

It is the single decision that separates askFinz from every other AI product: we read the web ourselves and keep what we find. Everything below follows from it — the price, the freshness, and what the answers can actually reach.

three modelsweb/
where answers come from/
 ├── traditional/
 │    ├── owns the index    yes
 │    ├── ends at           links
 │    └── built for         browsing
 ├── rented/
 │    ├── owns the index    no
 │    ├── ends at           an answer
 │    └── rate set by       someone else
 └── askfinz/
      ├── owns the index    yes
      ├── ends at           an answer
      └── rate set by       us
owned end to end2 of 3
Three ways to do this

Where the web knowledge comes from.

Where the web knowledge comes from

The traditional index

Google, Bing
The web
Their index
A page of links

Owned end to end, and built to send you somewhere rather than to answer you. It was designed decades before anything needed to read pages for meaning.

Returns links, not passages. Stores what a page links to rather than what it says, so an answer can point at a homepage but not at the paragraph that settles the question.

Everything is a web page. A judgment, a standard and a clinical trial are filed the same way as a blog post — no jurisdiction, no revision, no trial status to cite.

A snapshot, not a live picture. Crawling runs in cycles, so what you search is the web as it was when the cycle last reached it — days or weeks behind, and the same snapshot for everyone regardless of what anyone is actually asking about.

Refreshed on its own schedule. What gets re-read is decided by the crawler's plan, not by what people are asking today. Interest in a subject does not make it any fresher.

Ranked to sell placement. The order is shaped by an advertising business. That is a fine trade for browsing and a poor one for answering.

The rented index

Most AI products
Someone else's indexnot theirs to set
A fee per query
An answer
queries billed: 0

The quickest way to launch, and the reason it is the common choice. The chain starts outside the product, so the rate and the limits are set elsewhere.

The price is not theirs. Every answer carries a per-query fee set by the provider. A rate rise lands on the product's margin, not the provider's.

The ceiling is not theirs either. Rate limits, quotas and terms belong to whoever owns the index. Growth needs their permission as much as demand.

Reaches only what was sold. You get what the upstream index holds. Material it never collected cannot be bought back at any price.

It can end. When one large provider retired its search API in 2025, every product built on it had to move at once. That risk comes with the model.

Ours

askFinz
The web
We read it
Our index
An answer

Owned end to end and built to answer. Years of reading went into it, and that is the reason the price is ours to set rather than someone else's to raise.

Read continuously, and where it is wanted. Reading never stops and never waits for a cycle, and sources people actually ask about are re-read rather than left until the schedule comes round. What you search is the web now, not the web as it was.

Smaller than the giants. An index decades old is far larger. We compete on what is held, how it is filed and what it costs — not on being the biggest.

Ours to maintain. Storage, machines and quality control are our problem forever. Cheap per page is what makes that an ordinary cost rather than the business itself.

Two of these chains are owned end to end. Two of them finish with an answer. Only one does both — and that is the whole of why our pricing looks the way it does further down this page. Ours is not the largest index in the world and we do not claim to be; we compete on what is held, how it is filed and what it costs.

What changes as a result

Four differences you can check.

Read for meaning, not for links

A traditional index stores what a page links to. Ours stores what a page says, broken into passages that can each be found on their own — which is why an answer can point at the paragraph rather than the homepage.

Filed as what it is

A patent, a standard, a clinical trial and a judgment are not web pages. We keep each as the thing it is, with its identifiers and status — so a citation survives scrutiny.

Read from the publisher

Where an organisation publishes properly, we read it from them rather than scraping a worse copy off the open web. That is why our standards, patents and clinical collections are as complete as they are.

Fresh because it never stops

Traditional crawling revisits on a schedule and runs days to weeks behind. Ours reads continuously, and popular sources are re-read rather than read once and forgotten.

The economics

Cheap to hold is why it can be cheap to use.

Every page held costs storage forever. Keep that small and reading the whole web is a business; let it grow and the index eats the company. These are live figures.

4.4 KB
Per stored page
everything needed to find and rank it
1.7
Passages per page
each one separately searchable
47
Collections
each filed as the kind of thing it is
What a drive holds41.5× more per byte · 98% smaller
Capacity
askFinz243.8M pages

4.4 KB per page

CC — a public web archive, for reference5.9M pages

182.6 KB per page

The same 1 TB holds 243.8M pages here against 5.9M at the reference rate. That is the whole reason holding the web is an ordinary business cost for us rather than the business itself.

Our figure is read live from the running index. CC is a large public web archive whose scale is published openly — 364.01 TiB across 2.14 billion pages, which works out at 182.6 KB per page. It is not a like-for-like comparison and is not meant as one: an archive keeps fetched pages, which is a different job from keeping them searchable. It is used here because it is the one web-scale per-page storage figure anybody can check for themselves.

And not every page needs the same shelf.

Keeping everything on the fastest storage would price the index like a database. Newer material sits where answers are instant; older material moves to storage that is larger and cheaper per page — without ever leaving the index. A question about this morning and a question about five years ago are both answerable, and neither is charged as though it were the other.

Where the index lives
filings published todayanswered from Immediate
ImmediateThe newest of everything
fastest storage, smallest

Answers almost every search on its own.

RecentThe months behind that
larger, cheaper per page

Reached when a question looks further back than the newest material.

DeepYears of older material
largest, cheapest per page

Rarely needed — and the reason a thorough question still gets a real answer.

Sealed copyA second copy of everything
never on the search path

Held elsewhere, encrypted before it leaves us, and read only if something has to be rebuilt. It costs almost nothing precisely because nothing asks it questions.

Not every page deserves the same shelf. The newest material sits where answers are instant, and older material moves to storage that is larger and cheaper per page without ever leaving the index — so a question about last week and a question about five years ago are both answerable, and neither is priced as though it were the other. Times shown are measured live: the median of three timed requests from the search service to each shelf, refreshed every few minutes. They are the cost of reaching that shelf, not of a whole search.

What a rented answer can reach

A summary of page one is not an analysis of the web.

A search API hands back a ranked page of results — usually ten or twenty. Wrap those in a model and you get a fluent answer, but it can only ever be about the handful of pages that ranked. Everything below the cutoff is not weighed and discarded; it is never seen at all.

Ranking is not coverage

Those ten results were ordered for someone browsing — popularity, links, freshness, and in most cases a business that sells placement. None of that is a judgement about which pages answer your question, and a thorough answer needs the twentieth-best source as much as the first.

You cannot ask for everything

There is no query that means “every paper that measured this” or “all the judgments that applied this rule”. The interface returns a page of results, so the honest ceiling on any answer built from it is whatever fitted on that page.

Some of it was never collected

Registries, standards, filings and long documents are often not web pages in the shape a general crawler keeps. If the upstream index never held that material, no amount of paying for queries will retrieve it — the gap is in the supply, not the search.

A confident answer either way

This is what makes it hard to notice. The answer reads the same whether it drew on everything relevant or on the four pages that happened to rank. Nothing in the output tells you which, and the reader has no way to check.

Holding the index changes the question being asked. We are not requesting a page of results and summarising it — we search the passages themselves, across everything held, and rank for whether a passage answers the question rather than for whether a page deserves a click. When an answer says a study found the opposite, it is because the study is in there, not because it made the first page.

Ready to build on

One call returns the passage, not a list of places to go looking.

Anything that reads the web for you — an assistant, an agent, a research tool — needs the text, not the link. With a conventional search API that means two steps: ask for results, then fetch and strip each page yourself. Both cost money, both cost time, and the second one fails on exactly the pages that matter most.

Matched by meaning

Retrieval works on what a passage says rather than which words it contains, so a question phrased nothing like the source still finds it.

The passage comes back

Every result carries the relevant text with it — already cut to the part that answers the question, rather than a whole page to trim down afterwards.

No second round trip

No fetching each result separately, no extraction step, and no separate bill for page content. One request, and what comes back is ready to use.

This is why the prices further down compare the way they do. Where a provider charges once for a search and again for the page text, the rate shown is the one that includes both — because that is the call worth comparing to ours.

What a thousand searches costs

Our own index, so our own price.

ProviderPer 1,000 searches
askFinz — top-upadditional searches beyond your plan$0.75
askFinz — with the extensionincluded in what you already pay$1.00
askFinz — standalonesearch on its own$3.50
Keirosearch with page text, top tier; $5.00 and $9.00 below it. Ranked links alone $1.00$3.00
Brave Search APIanswers tier $4.00$5.00
Exa — answer$5.00
You.com Search API1–100 results per call$5.00
Linkup$0.005–0.006 per request across all tiers$5.00
Exa — standard searchup to 10 results per request$7.00
Tavilypay as you go, $0.008 per credit$8.00
SerpApientry plan; $9.17 at 30k/month$25.00
Exa — deep searchdeep reasoning $15.00$12.00

You will find cheaper rates than these. A separate group of services quotes as little as $0.60 per thousand, and they are worth understanding rather than dismissing: they don't hold an index of their own, they pass your query to a search engine and hand back what it returned. That is a genuinely useful thing to buy, and it is priced accordingly. It is also a different product from this one — it can only reach what the engine chose to show, and none of the material we hold because we read it from the source. Among services that run their own index, the rates above are the market.

The budget option is gone. Bing's Search API — what much of this market was built on — was retired in August 2025. Its replacement is a different kind of product at a considerably higher price, which is part of why search costs what it does elsewhere.

Rates read from each provider's own pricing page on 12 August 2026. Where a provider gets cheaper with volume, the comparison uses their best rate, not their entry price. Rates are matched on what the call returns: askFinz search returns the relevant passage from the page with every result, so where a provider charges separately for page text, that is the rate shown. Providers change pricing — check theirs before relying on this.

Questions

Isn't building your own index wildly expensive?

It is expensive to start and cheap to run. Storage is what decides that, and ours costs a few kilobytes per page — small enough that holding the whole thing is an ordinary business cost rather than the business itself.

Why does it matter to me who owns the index?

Because it sets your price and your ceiling. A product built on a rented index passes on every price rise and inherits every rate limit. When one large provider closed its search API in 2025, everything built on it had to move overnight.

Can I buy access to the index directly?

Not today. It powers askFinz Search and the apps built on it rather than being sold as a feed. That is a decision rather than a limitation — if there is enough industry interest, it is something we would look at.

Check it yourself

Every claim here is testable.

The rates came off each provider's own pricing page and carry the date they were read. The storage and quality figures come from the running system, not a slide. Search it, watch it work, or read how it is built — whichever you would need to see before believing any of it.

askFinz vs traditional search indexes — askFinz