Best Amazon Scraping APIs in 2026: Field-by-Field

Amazon data APIs are priced on the page and understood in the invoice. Almost every comparison post ranks the vendor that wrote it, pairs a headline per-1,000 rate with an unspecified field set, and never shows what the price actually buys. This one does the arithmetic the way a developer paying the bill would: normalize every provider to per-1,000 for the same job, list the fields you get at that price, and show what each API reports when it cannot return everything.

The method and every competitor price below are dated. This post was first published on 25 August 2026; every competitor rate in it was re-read on each vendor's own live pricing page on 8 September 2026 and several had moved, so the tables below are the September figures. I present Serpent (the API this site sells) as one option among several, no further. First-person "I" here means my research method — the sources and the normalization — not a pitch.

How I compared

I judged providers on the three things that decide a data pipeline: the true cost per 1,000 at the tier you will actually use, the number of fields you get for that price, and what the API says when it cannot return the full result. I normalized every provider that prices per result or per credit down to per 1,000, and I opened each vendor's own pricing page on 8 September 2026.

Two rules keep the cost table honest, and they are worth stating before you read it. First, if I show our lowest rate I have to show every rival's lowest rate too — comparing our best price against someone else's entry price is the oldest trick in this genre. Second, a floor is only real if you can say what it costs to stand on. So the table carries two rates per provider: the price on the cheapest plan you can actually buy, and the lowest price the vendor publishes anywhere, with the plan that unlocks it named. Several of those floors need $999 to $30,000 a month in committed spend. Where a printed rate sits behind a quote-only enterprise tier rather than a plan with a price on it, I say so, because "starts from" is not the same promise as a published plan. All prices are monthly-billed unless noted; Rainforest and ScraperAPI publish annual discounts (−20% and −10%) on top.

I did not publish an accuracy percentage, and I will not. A synthetic accuracy figure says nothing about the specific page you are parsing, and it is the number I distrust most from vendors who have the most to prove. What you can verify yourself with one call is the field set and the response honesty, so that is what I measured.

Cost per 1,000, normalized

The table below is the same arithmetic applied to each published rate. Entry is the price on the cheapest plan a new customer can actually buy. Lowest published is the cheapest rate the vendor prints anywhere on its own card, and the last column names what you have to be paying to get it — because a $0.027 rate behind a $30,000-a-month commitment is not the same product as a $0.027 rate you can buy today.

Every competitor figure above was read on that vendor's own live pricing page on 8 September 2026: apify.com/pro100chok/amazon-scraper and apify.com/pricing for the store-discount tiers, oxylabs.io/products/scraper-api/web/pricings (Amazon has its own published rate there, distinct from the Google and general-web rates), scraperapi.com/pricing with the credit cost from its credits documentation, trajectdata.com for Rainforest, brightdata.com, scrapingdog.com/pricing, and serpapi.com/pricing. Serpent's rates are its own published card. Note the unit differs across the row: Apify, Oxylabs and Bright Data bill per result row; ScraperAPI, Scrapingdog, Rainforest, SerpApi and Serpent bill per request. That matters most on search, where one request returns a page of rows.

What the price actually buys

The price is only worth comparing next to the field list. On Serpent, one product call returns price and list_price as numbers, ISO-4217 currency, a numeric rating, ratings_total, a small labelled reviews sample plus a separate reviews_count, the rating_histogram as {stars, percentage} rows, features, category, rank, availability, variants_count, image_count, media and videos_present. Search returns a ranked results[] with sponsored and prime flags; bestsellers returns category, category_slug and its ranked list.

A low rate often buys a smaller card. A $0.20-per-1,000 rate can mean a basic search row with no review count, no histogram, no variant or media metadata. That is not a defect; it is a scope difference. It only becomes a problem when you compare two $0.20-per-1,000 cards thinking they are the same item.

The other thing to read before you compare any two numbers is the unit. Apify, Oxylabs and Bright Data charge per row returned: a search that brings back 30 product cards is 30 billable records. Scrapingdog, ScraperAPI, Rainforest, SerpApi and Serpent charge per request, and a Serpent search request returns a whole page of roughly 16 to 48 cards for one charge. On a product lookup the two units coincide, because one product page is one record either way. On search they do not, and a per-row vendor that looks cheaper per 1,000 can be an order of magnitude more expensive for the same crawl. Neither unit is dishonest; reading the wrong one is what costs money.

The 2026 review ceiling

This is the thing most Amazon API posts get silently wrong: reviews are no longer fully reachable. As of 2026 Amazon restricts public review browsing — the review pages require sign-in and past the first page they return the same content again. The resulting ceiling for any provider is the small set of reviews Amazon renders on the product page itself (about a dozen for a typical listing). Vendors that frame "reviews" as a full crawl are describing pre-2026 behaviour or a workaround that trades completeness for stability.

That is why the distinction between a review sample and a review count matters. The count is a single number that stays accurate; the sample is the handful of rows you can actually get. Building a pipeline that treats the two as the same is how an analysis quietly undercounts sentiment.

Review-browsing restriction documented in the Apify Amazon scraper's own field notes (Aug 2026); the sample-vs-count distinction is a field convention I measured across the providers above.

Under-delivery, made visible

An API is only as good as what it tells you it did not return. This is the axis almost no comparison checks, and it is where a pipeline either stays honest or quietly builds a dataset with holes. On Serpent an Amazon page arrives whole or not at all, so there is no silently truncated record to detect: a call that runs out of time says so and returns an async_endpoint naming where to send it, a failure on our side is refunded in full, and a page that genuinely holds nothing is an answer and is charged. Search beyond the synchronous page cap returns a 400 that names the async form instead of silently truncating.

None of that is a claim that other providers are worse. It is a different contract: some vendors return a clean array and leave the gap for you to find in your own analysis. Neither approach is wrong; the difference is whether the gap is discoverable from the response or only from your results.

A real cost model: 10,000 products

To make the sticker prices concrete, I ran the same workload across the published rates: collect 10,000 full product records a month for a price monitor. This is where per-1,000 rates stop being the answer, because most of these vendors have a plan minimum well above the value of the work. The column that matters is what actually leaves your account in a month.

Model method: 10,000 full product records per month, one request each, priced at the smallest plan on each vendor's live card that covers the volume — read 8 Sep 2026. Where a vendor sells usage without a subscription (Serpent, Bright Data) the bill is the usage. This is a data-collection cost only; it excludes the engineering time you spend parsing and maintaining the response shape, and it excludes overage rates, which every subscription vendor above charges separately.

Two things fall out of that table, and the first one is not about the rate at all. At 10,000 records a month you are mostly buying a plan, not data: Scrapingdog bills $40 for $2.00 of work and Oxylabs $49 for $5.00 of it, so at this volume the subscription writes the invoice and the per-1,000 rate barely enters it. That reverses as you scale, which is the point at which the rate starts to matter: at a million records a month Scrapingdog's Standard plan is $90 against $14.00 for the same work on our Scale rate. The second is that among the seven rivals here the published price for the same field set spans $0.027 to $25.00 per 1,000, and the low end of that range is a $30,000 monthly commitment. Find the volume you will actually run before you compare a single rate, then read what the record contains, then read what the API tells you when it under-delivers.

The providers, fairly

1. Serpent. Product, search and bestsellers across 22 marketplaces. Numeric price and list price, ISO-4217 currency, numeric rating, aggregate thumbs, a labelled review sample with a separate count, a rating histogram, features, category, rank, availability and media metadata. Billed per request from a deposited balance with no monthly plan, so at small volume the bill is the usage and nothing else, and the rate — $0.020 per 1,000, $0.014 at Scale — is below every rate published in this table. Prelaunch.
2. Apify — pro100chok/amazon-scraper. Five billable record types in one actor — search/category rows, product detail, reviews, seller profiles and offers — each priced separately, so the bill moves with every record type you add. Apify is a marketplace rather than a single vendor, and the Amazon actors on it do not price alike: junglee/amazon-crawler, the most-used one, publishes "from $3.00 / 1,000 results" and charges $5.00 per 1,000 on the free plan. Check the actor, not the platform.
3. Oxylabs Web Scraper API. An enterprise-oriented per-result API with a large proxy estate; strong when Amazon is one target among several. Worth knowing that Amazon is its cheapest published target — $0.50 per 1,000 where Google is $1.00 and general web is $1.15 — so an Oxylabs quote read off the generic rate overstates the Amazon cost.
4. ScraperAPI. A general scraping API where Amazon costs 5 credits a request against 1 for a standard page. Predictable credit plans and an active ecosystem; the credit multiplier is what puts its entry rate near the top of the table even though its credits are cheap. Note also that these are no longer two independent quotes: Rainforest's parent, Traject Data, carries a "Traject Data joins ScraperAPI" banner across its site as of this reading.
5. Rainforest API. The most complete field coverage for offers, sellers and buy-box data, built for heavy, consistent volume, and the only vendor here that is Amazon-first rather than Amazon-among-many. Subscription priced at 1 credit per request, so it suits a steady workload rather than a burst — and its entry plan is the most expensive entry point in the table apart from SerpApi.
6. Bright Data. The deepest data catalogue and the biggest network, billed per successful record. The one rival here besides us with no plan minimum: $1.50 per 1,000 pay-as-you-go plus a 5,000-record monthly free tier, which makes it the easiest of the large vendors to trial honestly.
7. Scrapingdog. A credit-based Amazon product, search and bestsellers scraper, and the cheapest per request of the rivals here — Amazon Product and Amazon Search are 1 credit each, which is $0.20 per 1,000 on the $40/month Lite plan. Its $0.027 floor is the lowest rate any other vendor in this table publishes, and it is printed on a $30,000-a-month plan. Bestsellers costs 5 credits. One subscription covers its whole API catalogue, and the cost of a call is the plan you are on, so its economics only work in your favour once your volume fills the plan you bought.
8. SerpApi. Primarily a search engine result API; Amazon sits on its own engine billed as one search. If you already use SerpApi for SERP, it blends into a bill you are already paying; as a dedicated Amazon source it is the most expensive entry point here by a wide margin.

Which one to pick

Choose on the volume you will actually run, then on the data you must have, then on the honesty contract. Nobody wins every row of the table above, and a post that says otherwise is selling something.

• You want one subscription covering far more than Amazon, and the volume to fill it — Scrapingdog, at 1 credit per Amazon Product or Search call, where the plan you buy covers its whole API catalogue. Its per-request rate is the lowest here after ours, and once your volume fills the plan the unused remainder stops being the problem it is at 10,000 records.
• You are running a few thousand records a month and do not want a subscription — Bright Data pay-as-you-go, or Serpent. These are the only two here with no monthly minimum, so the bill is the usage and nothing else.
• You need seller / buy-box / offer depth at steady volume — Rainforest or Oxylabs.
• You want one Amazon actor with multiple record types and per-event billing — an Apify Amazon actor, chosen by reading that actor's own pricing tab rather than the platform's.
• Amazon is part of a much broader data estate — Bright Data or Oxylabs, where the Amazon rate rides on infrastructure you are buying anyway.
• You want a full product record with a separate review count, a flat per-request price, and under-delivery reported in the response — put Serpent, with its documented field set, on your shortlist. It is prelaunch, and the rate card is published so you can plan.

Before you commit on any of them, run one live call on a product you care about and read the response the way I would: check the field list, the review count versus the sample, and whether the API would tell you if it had to drop rows. The endpoint reference and the Python walkthrough show the shapes so you can compare apples to apples.

FAQ

Is any Amazon API still able to return every review on a product?

Not in practice. Amazon now restricts public review browsing: the review pages ask for sign-in and past the first page repeat the same content, so the available set at any provider is the small review sample rendered on the product page. Compare a labelled sample with a separate review count rather than assuming every review is reachable.

How should I compare Amazon scraping APIs on cost?

Normalise every provider to a per-1,000 price for the same work, then check three things the headline rate hides. First, the plan the rate needs: across the seven rivals compared here the lowest published rates sit on plans costing $999 to $30,000 a month, so a floor you cannot reach is not a price. Second, the plan minimum at your volume: at 10,000 records a month most of these vendors bill their subscription rather than your usage. Third, the unit — some vendors charge per row returned and some per request, and on a search call those differ by the size of the page.

Which Amazon scraping API is cheapest?

It depends on what you count. On price per request Serpent is the lowest published rate in this comparison: every Amazon endpoint is $0.020 per 1,000 requests with no plan fee at all, against Scrapingdog's $0.20 per 1,000 on its $40-a-month Lite plan and its $0.027 floor, which needs a plan costing $30,000 a month. On the total monthly bill the order changes again, because most vendors here bill a subscription rather than your usage: at 10,000 records a month Scrapingdog charges $40 for $2 of work. Work out the volume you will genuinely run before you compare a single rate. Prices read 8 September 2026.

Which Amazon marketplaces are commonly covered?

Coverage varies. Many providers cover the big English and EU storefronts (com, co.uk, de, fr, co.jp, in, com.au, ca). Serpent's own product, search and bestsellers endpoints cover 22 marketplaces via a domain or country parameter.

What does a structured product record typically include?

A full record usually carries a numeric price and list price, ISO-4217 currency, a numeric rating, an aggregate ratings total, a labelled review sample and a separate review count, a rating histogram, features, category, rank, availability, variant count, image count and media flags. The exact set is what you should compare, because it varies more than the price.

Is the Serpent Amazon API live today?

It is in prelaunch. The endpoints and the rate card below are documented now so you can plan, and the page tells you how to be notified when they ship.

What happens when an Amazon API returns nothing?

Behaviour differs by provider. On Serpent an Amazon page that genuinely returns nothing is an answer and is charged, while a failure on our side is refunded in full. The Amazon endpoints deliver a page whole rather than in parts, so instead of a half-built record you get a call that says plainly it ran out of time and an async_endpoint field naming where to send it.