Best X (Twitter) API Providers in 2026 (Real Prices)

Every comparison of X (Twitter) API providers written before 2026 is now wrong about the most important number on the page. X did not raise its prices this year — it deleted the price list. The $200 Basic plan and the $5,000 Pro plan are being retired, the free tier is gone as a self-serve option, and what replaced them is a credit meter that bills you per record returned.

That single change is why the third-party market for X data exists, and why picking one is harder than reading a headline rate. This post is a directory with the numbers attached: who the real providers are today, what each one actually sells, what it costs at list price, and who each one suits.

How I compared

This is desk research, and I want to be exact about what that means. I read every rate below off the vendor's own live pricing page on 9 September 2026 and linked the page I read it from. I went hands-on with the documentation, the pricing calculators and the free-tier terms — not with a benchmark. I ran no latency test and no success-rate test, so you will find no latency or block-rate claims here. Anyone publishing those numbers for eight providers at once owes you a method; I would rather publish the arithmetic I can show.

I judged each provider on three things a buyer can actually check before signing up: the unit they bill in, the published price in that unit, and the scope — whether you are buying lookups, keyword search, or a full data feed. Those three decide the invoice. I also note the free tier or trial in every case, because the size of the thing you can test with tells you how much of the risk the vendor is willing to hold.

I sell one of the options in this table, so treat the Serpent rows as a vendor quoting its own rate card. Everything about it is checkable on the public pricing page and the API reference, and nothing in the comparison depends on taking my word for it.

The official X API in 2026: the tiers are gone

Start here, because it is the fact the rest of the market is priced against.

X began testing pay-per-use in a closed beta in October 2025 and made it the default on 6 February 2026. The flat tiers did not survive the year. Legacy Basic plans were deprecated and migrated to pay-per-use after 1 June 2026, and legacy Pro plans followed after 1 September 2026. Today X's own introduction page says it plainly: "The X API uses pay-per-usage pricing. No subscriptions — pay only for what you use."

If you land on a 2025 guide listing Free, Basic, Pro and Enterprise with monthly prices, that page is describing a product you can no longer buy.

The published rates

Source: X API pay-per-usage pricing and credits, read 9 September 2026. Re-check before you budget — X's page states rates are subject to change and the Developer Console is authoritative.

Three clauses in that pricing page matter more than the numbers, and most write-ups skip all three.

• Reads are "charged per resource returned in the response." The billable unit is a record, not a request. A search returning twenty posts bills twenty Post reads.
• There is a hard ceiling. "Pay-per-usage plans are capped at 3 million Post reads per monthly billing cycle. If you need higher volume, upgrade to an Enterprise plan." Above the cap the answer is a sales conversation, not a bigger invoice.
• Repeats inside a day are free. "All resources are deduplicated within a 24-hour UTC day window" — request the same Post twice in that window and you are charged once. If you poll a fixed set of posts, that clause is worth real money and it is the one genuinely generous thing in the new model.

The nearest thing to a free tier is a rebate rather than an allowance: X offers "up to 20% back" in xAI API credits based on X API credit spend. Useful if you were buying xAI credits anyway; not a way to evaluate the API for nothing.

Per post vs per 1,000: what the unit quietly hides

Half this market quotes a price per tweet returned and half quotes a price per request made. Both get written as "$X per 1,000", and the two numbers mean completely different things.

On a single-post lookup the units coincide: one request, one record, so a per-item rate and a per-call rate are directly comparable and the cheaper number simply wins. On anything that returns several records at once they diverge fast, and the gap grows with the depth of what you collect.

Worked example. Say you want one profile plus its five most recent posts, with engagement on each. On a per-resource biller that is six billable records. On a per-call biller it is one charge. At X's own rates that basket is one user read plus five Post reads — $0.010 + 5 × $0.005 = $0.035, or $35.00 per 1,000 profiles. The same basket on a $0.15-per-1,000-items vendor is roughly $0.90 per 1,000. On a per-call vendor it is the price of one call.

None of that makes the per-item vendors expensive — they are 30 to 40 times cheaper than the official API on the same basket. It makes the headline number insufficient. Before you compare two rates, write down how many records your typical job returns and multiply. This is the same trap I mapped for TikTok providers, where the sticker price and the real price diverge by depth rather than by unit: see the cost-per-depth ladder for TikTok scraping APIs. It is also why the costs a headline rate hides is worth reading before any procurement decision.

The price table — prices checked 2026-09-09

Every figure below is the vendor's own published rate, read from the linked page on 9 September 2026. The unit column comes first on purpose: it is worth more than the price column.

Prices checked 2026-09-09, each from the vendor's own linked page. GBP converted at 1 GBP = 1.3539 USD (rate of 8 Sep 2026); Data365 publishes in euros and I have left it there. ScrapingDog's per-1,000 figures are derived: its X profile documentation states each successful request costs 5 API credits, so the Lite plan's 200,000 credits buy 40,000 requests. ScrapeCreators and SocialCrawl per-1,000 figures assume a one-credit endpoint. Vendors change prices without notice — re-verify anything you are about to sign for.

The providers, one by one

1. X API (official). The only source with a licence to the firehose and the only one that can write. Nothing else on this list can post, DM, or read your own app's data at the discounted owned-read rate. It is also the only one that bills per resource returned, which is what makes it tens to hundreds of times the cost of the alternatives on a read-heavy job. Buy it if you write to X, need contractual data rights, or your compliance team requires a first-party agreement.

2. twitterapi.io. The clearest per-item price list in the category: $0.15 per 1,000 tweets, $0.18 per 1,000 users, followers from $0.01 per 1,000 on a tiered scale, and a stated minimum charge of $0.00015 per API call. Broad coverage — search, timelines, followers. On Trustpilot it holds 4.6 out of 5 from 30 reviews, with support responsiveness the most-praised theme and one December 2025 reviewer reporting the opposite experience. Thirty reviews is a small sample; read them yourself rather than trusting the average. Buy it if you need breadth across search, timelines and follower graphs in one account.

3. Apify actors. Not one product but a marketplace, and the variance between actors is larger than the variance between vendors. Published tweet rates on 9 September ran from $0.15 per 1,000 results (igolaizola) and $0.15 per 1,000 delivered rows (xquik) up through $0.18 (kaitoeasyapi) to from $0.40 (apidojo Tweet Scraper V2, 98K users). A per-query variant, Twitter Scraper Unlimited, bills $0.016 per query with the first two pages (about 40 tweets) included, which is a genuinely different shape worth modelling separately. Apify holds 4.8 out of 5 from 727 Trustpilot reviews; the recurring complaint in the negative ones is cost per 1,000 data points on specific actors rather than reliability. Buy it if you want maximum feature coverage and are comfortable evaluating individual actors, including who maintains them.

4. Bright Data. The enterprise end. $1.50 per 1,000 records pay-as-you-go, $1.30 per 1,000 on the $499-per-month plan, and a free tier of 5,000 records a month with no card — by a distance the largest free allowance here. Its page commits to "pay only for what's successfully delivered." Trustpilot has it at 4.3 out of 5 from 1,019 reviews, a wide spread in which support quality is both the loudest praise and the loudest complaint. Buy it if you need procurement paperwork, a named contact and a large trial more than you need the lowest unit price.

5. ScrapeCreators. Flat credit packs across 37+ social APIs: $47 for 25,000 credits ($1.88 per 1,000) or $497 for 500,000 ($0.99 per 1,000), 100 free credits to start, and credits that never expire. Most endpoints cost one credit; some cost more, and the docs state which. Buy it if you want X alongside several other networks on one key and value credits that do not evaporate at the end of the month — a real problem elsewhere, as the survey of credit-expiry rules shows.

6. ScrapingDog. Monthly credit plans from $40 (200,000 credits) to $350 (6,000,000), with 200 free credits to test. Each X request costs 5 credits, so the effective rate runs from $1.00 to $0.29 per 1,000 requests. Per-request billing rather than per-item, which is the right unit for lookup work — the trade is a monthly commitment. Buy it if you already use ScrapingDog for other targets and want X on the same credit balance.

7. SocialCrawl. A very wide catalogue — it advertises 572 APIs across 65 platforms — on non-expiring credit packs from £15. Standard endpoints cost one credit, mixed endpoints three, and universal search twenty, so your effective rate depends heavily on which endpoints you call. Buy it if breadth across many networks matters more than the per-call price on any one of them.

8. Data365. A monthly-commitment social data API: €300 for 500,000 credits on one network, €850 for 1,000,000 across three, custom above that, with a 14-day trial. Note the honest gap — the public page does not state how many credits an X request consumes, so you cannot derive a per-record cost from it without asking. Buy it if you are buying several networks on a contract and will get a quote anyway.

9. Serpent. Deliberately the narrowest product here: two lookups. One post by id or permalink, and one public profile by handle with the recent posts that profile shows. Billed per call at $0.10 per 1,000 for either, with hydrate=true adding engagement and media to each recent post at 1.5× the base ($0.15 per 1,000, flat at every tier). Ten free API calls, no card, no monthly plan and no daily bucket. What it does not do is published on the product page rather than discovered on your invoice: no keyword search, no full timelines, no follower or following lists, and no retweet, quote, bookmark or view counts. Buy it if you already hold ids or handles and want the lowest per-record cost with the fewest surprises.

Two names people expect that are not here: RapidAPI wrappers, which churn too fast for a dated price to stay true for a week, and the free Python libraries, which the login wall below took out of contention.

One basket, priced five ways

This is the table I could not find assembled anywhere, and it is the whole argument of the post in one block. The basket is 1,000 profiles, each with its five most recent posts and the engagement on them. Five is my own modelling assumption, not a promise about how many recent posts any given profile shows — change the number and the ranking holds, only the gaps move.

Arithmetic, not measurement. Official X API: $0.010 + 5 × $0.005 = $0.035 per basket. twitterapi.io: $0.00018 + 5 × $0.00015 = $0.00093. Apify xquik and Bright Data: 6 records × the published per-record rate. Serpent: one hydrate=true profile call at its flat $0.15 per 1,000. The credit-based vendors are absent because their published pages do not state credits per X endpoint in a form I could convert without guessing — ScrapingDog is the exception at 5 credits per request, but its profile response scope is not stated clearly enough to call it the same basket.

The honest reading of that table is not "one provider wins". It is that the unit, not the rate, is doing most of the work. Per-call billing wins on lookups where one call covers a whole record; per-item billing wins nothing here but is the only sane unit for keyword search, where you genuinely do not know in advance how many records come back. Pick the unit that matches the job, then compare prices inside that unit.

Caps, rate limits, and what you cannot get without a login

Three constraints shape this category and none of them is a price.

The official cap is a ceiling, not a bill. Pay-per-usage stops at 3 million Post reads per monthly billing cycle. There is no "pay more and keep going" — the documented path above that number is Enterprise, which is a sales conversation rather than a checkout page — TechCrunch put the pre-change Enterprise tier at $42,000 a month. Model your peak month against that ceiling before you build on the official API.

The login wall is why free scraping stopped working. In mid-2023 X began requiring people to log in to view posts and profiles, redirecting logged-out visitors to a sign-up page, alongside daily read limits Elon Musk described as a temporary emergency measure against data scraping. Search and full timelines have needed an account ever since. Every tutorial that fetches a public X page with two lines of Python dates from before that change. This is the single biggest reason the third-party market exists at all, and it is why "just build it yourself" is a bigger project in 2026 than it looks.

Rate limits are published unevenly. The official API documents per-endpoint limits alongside the credit meter; the marketplace actors mostly express throughput as concurrency on your plan; ScrapingDog states concurrency explicitly per plan (5 on Lite up to 150 on Premium). Serpent publishes no daily bucket and no monthly minimum. If sustained throughput matters, ask for the number in writing before you commit — and read how documented rate-limit caps compare across data APIs for what "unlimited" usually means in practice.

One more dimension nobody advertises: what a failure costs you. Per-item vendors that bill only for delivered records make an empty result free by construction. Per-call vendors have to make a policy choice and state it. Serpent's is on the record — a missing post or handle is a charged answer, and anything that is our problem rather than the resource's is refunded — and the general shape of that decision is worth understanding before you sign anywhere: who pays for a failed request.

Which one to pick

Decide the unit and the scope before you look at a rate.

• You need to write to X, or you need contractual data rights. The official API, and there is no second option. Budget against the 3 million Post read cap and use the 24-hour deduplication window — on a polling workload it is real money.
• You need keyword discovery across the whole platform. twitterapi.io or an Apify search actor. Budget per result returned, not per call, and cap your queries.
• You need one key across many social networks. ScrapeCreators, SocialCrawl or Data365. Compare them on credits-per-endpoint, not on the headline pack price.
• You need procurement paperwork and a large trial. Bright Data. 5,000 free records a month is the biggest evaluation allowance in this list by an order of magnitude.
• You already hold post ids or handles and want the lowest cost per record. Serpent's X API at $0.10 per 1,000 calls, with ten free calls to test. Check the scope list first — if you need search, timelines or follower graphs, it is the wrong product and I would rather you knew now.

Whichever you pick, do the same two things before you commit. Pull one real record on the free tier and read the whole response, not just the field you came for — that is when you find out which counts are missing and which ones arrive as a permanent null. And write your integration against a normalised internal shape rather than one vendor's JSON, because you will change vendors in this category: normalising JSON across multiple providers is a two-hour job that saves a rewrite.

Same buying question on a different network? The LinkedIn data API comparison covers a market with the same unit confusion, much harder terms, and the same official-versus-third-party choice this post opens with. The social API index lists what else is covered, including the Reddit API. Before you build anything at volume, the legal and ethical ground rules for collecting search and social data is the section most buyers skip and later wish they had not.

FAQ

How much does the official X API cost in 2026?

X moved to pay-per-usage pricing. Its published read rates are $0.005 per Post, $0.010 per user, $0.010 per following or followers resource and $0.001 per like, charged per resource returned in the response. Reads of your own app's data are $0.001 per resource. Writes are $0.015 per Post created and $0.200 when the Post contains a URL. Pay-per-usage is capped at 3 million Post reads per monthly billing cycle, and above that X directs you to an Enterprise plan. Rates read from the X API pricing page on 9 September 2026.

Is there still a free X API tier?

Not as a self-serve plan. X announced pay-per-usage on 6 February 2026 and made it the default for new developers. Legacy Basic subscribers were moved onto it after 1 June 2026 and legacy Pro subscribers after 1 September 2026, so what remains self-serve is pay-per-usage, with Enterprise above it. The nearest thing to free is a rebate rather than an allowance: X offers up to 20 percent back in xAI API credits based on what you spend on X API credits.

What does charged per resource returned actually mean?

It means the billable unit is a record, not a request. X's pricing page states that reads are charged per resource returned in the response, so a search that comes back with twenty posts bills twenty Post reads rather than one call. That is why a price per 1,000 tweets and a price per 1,000 calls are not comparable numbers until you know how many records a call returns.

What is the cheapest published rate for a single X post lookup?

Of the rates I could read on a vendor's own page on 9 September 2026, Serpent's $0.10 per 1,000 calls is the lowest, ahead of twitterapi.io at $0.15 per 1,000 tweets and the cheapest Apify tweet actors from a published $0.13 per 1,000 results. One post is one item on either unit, so that comparison needs no conversion. Two caveats belong with it: Apify's store discount can take an actor below its published floor and the discounted number is not published, and ScrapingDog reaches $0.29 per 1,000 requests on its $350 per month plan.

Can any of these providers search X by keyword?

Some can and some cannot, and it is the first thing to check. Apify's tweet actors, twitterapi.io and Bright Data all sell keyword and advanced search. Serpent does not: it sells two lookups, one post by id or URL and one public profile by handle, with no search parameter, no full timelines and no follower lists. If discovery is the job, buy it from a provider that has it rather than hoping a lookup product will stretch.

Which X data provider should I pick?

Pick by the unit you will be billed in, not by the headline number. If you already hold post ids or handles and want the lowest cost per record, a per-call provider wins because one call covers a whole record. If you need keyword discovery, a per-result search actor is the right shape and you should budget by results returned rather than by calls made. If you need contractual guarantees, compliance paperwork and an account manager, that is what Enterprise pricing and the larger data vendors exist for.