Retrieval and data-API vendors widen free tiers to grab evaluation
Retrieval, scraping, extraction, and vector-DB API vendors are competing on the width of the free/evaluation entry — enlarging recurring free tiers and cutting entry floors to win developer land while planning to monetize at production volume. In one 2026-07-14 window, Bright Data added a recurring 5,000-results/month no-card free tier, Exa 20×'d its free tier (1,000 → 20,000 requests/month), Powerdrill put a 1,000 daily-refreshed-credit pool on every tier including Free, Unstructured turned its 15,000 free pages into a monthly reset (and capped PAYG bills), and turbopuffer cut its entry-plan minimum 75% ($64 → $16). None touched their core per-unit rates — and the same vendors gate their premium/intelligence surfaces upward, so the pattern is a barbell.
What's happening — and why
What's happening: five retrieval/data-API vendors moved the same week (2026-07-14) to lower the barrier to evaluation. Bright Data added a genuinely recurring 5,000-results/month no-card free tier to its Web Unlocker / SERP / Web Scraper APIs, replacing a one-week 1,000-record trial. Exa raised its free tier 20-fold, from 1,000 to 20,000 requests/month, while leaving its per-1k-request endpoint rates unchanged. Powerdrill re-packaged its Bloom credit model so every tier — Free included — now carries a 1,000 daily-refreshed-credit pool. Unstructured turned its 15,000 free document pages from a one-time allowance into a monthly reset and added a Pay-As-You-Go bill cap above which pages are free up to 1M/month. And turbopuffer cut its entry-tier Launch-plan minimum 75%, from $64 to $16. None of them cut their core per-unit rates — the competition is on the width of the free/eval entry, not the marginal price.
Why: for usage-metered data and retrieval infrastructure the marginal cost of a small evaluation workload is near zero, so a generous recurring free tier is cheap customer acquisition — the revenue arrives at production volume. The countervailing move is that the SAME vendors gate their premium and intelligence products upward: Bright Data raised its Bright Insights eCommerce-intelligence list 5–8× ($250/mo and $400/mo → $2,000/mo each) the same day its scraping APIs gained a free tier, and Linkup's new closed-beta Extract endpoint requires a $10 minimum account balance rather than a free allotment. So the shape is a barbell — commodity request meters open to a free land-grab while high-value intelligence surfaces gate up.
How it works
Evidence over time
15 supporting · 11 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Bright Data | Jul 2026 | Unified Web Unlocker, SERP API, and Web Scraper API onto a Free / PAYG / Scale / Enterprise ladder and added a genuinely recurring free tier — 5,000 results or records per month, no credit card — replacing the previous one-week, 1,000-record trial. A recurring free tier as an evaluation on-ramp on the request-metered products. |
| Exa AI | Jul 2026 | Raised its free tier 20x, from 1,000 to 20,000 requests per month, while leaving core per-1k-request endpoint rates (Search, Deep Search, Contents, Answer) unchanged. Explicitly framed as lowering the barrier to evaluation. |
| Powerdrill | Jul 2026 | Re-packaged its Bloom credit model so every tier — Free included — now carries a 1,000 daily-refreshed-credit pool; the Free tier gained 1,000 daily credits where it previously had no allowance. A recurring (daily) free allotment added to widen the top of the funnel, headline monthly prices unchanged. |
| Unstructured | Jul 2026 | Turned its 15,000 free document pages from a one-time allowance into a monthly reset (a recurring allowance that can run a modest pipeline indefinitely at no cost) and added a Pay-As-You-Go monthly bill cap above which pages are free up to 1M/month — lowering both the entry and the scale barrier on the per-page meter. |
| turbopuffer | Jul 2026 | Cut the entry-tier Launch plan monthly minimum from $64 to $16 (75% cut, effective July 1 invoices), lowering its cheapest paid floor to narrow the gap with free-tier vector-DB rivals — though turbopuffer still offers no free tier. Per-unit usage rates unchanged. |
| You.com | Jul 2026 | New adopter on a flagship endpoint: the Web Search API pricing card, which previously led with a flat '$5.00 / 1k calls' and no daily allowance beyond the general $100 account credit, now leads with 'Free / 100 calls per day' and applies the $5.00/1k rate only above that quota — roughly 3,000 free calls a month. Every other endpoint held (Contents $1.00/1k pages, Research lite $12 → exhaustive $450, Frontier $1,200, Finance deep $110 / exhaustive $500). The change has not yet reached You.com's own Search API docs, which still describe the flat rate. |
| Firecrawl | Jul 2026 | Bundled a flat +1,000 search credits per month into EVERY plan card — Free, Hobby, Standard, Growth, Scale and custom Enterprise all get the identical allotment, shown as its own line beneath each tier's headline credit pool, on top of the unchanged 2-credits-per-10-results Search rate. Timed with a new banner ('Introducing our most accurate /search yet'), so a quality improvement shipped with a free allowance rather than a price. |
| Exa AI | Jul 2026 | Second widening in 15 days, and a unit change: signup credit doubled $10 → $20, the recurring monthly credit rose $7 → $10 and no longer requires a payment method on file, and the explicit '20,000 requests per month' Free Tier framing was dropped from both the pricing page and the billing docs in favour of pure dollar credits. Paid endpoint rates unchanged (Search $7/1k, Deep Search $12-15/1k, Contents $1/1k pages, Monitors $15/1k, Answer $5/1k). |
| Socket | Jul 2026 | 10x'd the Free plan's tracked-dependency ceiling from 1,000 to 10,000 while Team ($25/developer/mo) and Business ($50/developer/mo) list prices and the custom Enterprise tier held steady (confirmed via Wayback captures dated 2026-06 and 2026-07, the live page sitting behind a Cloudflare bot wall). A dependency-data developer tool widening the free evaluation envelope by an order of magnitude. |
| Apify | Jul 2026 | Doubled per-Actor RAM ceilings on every tier at unchanged prices — Free 8 → 16 GB, Starter/Creator 32 → 64 GB, Scale 128 → 256 GB, Business 256 → 512 GB — with monthly prices ($0/$29/$199/$999), compute-unit rates ($0.2 → $0.13/CU), concurrency limits and proxy/storage rates all held. The free tier's capacity, not its price, is the competitive variable on a scraping platform. |
| LlamaIndex | Jul 2026 | Flattened seat entitlements to 100 users on Free, Starter AND Pro (plan cards previously 1 / 5 / 10; compare table 1 / 10 / 25), leaving prices ($0 / $50 / $500), included credits (10K / 40K / 400K), PAYG caps ($500 / $5,000) and the 1,000-credits-for-$1.25 rate untouched. Removing the seat gate from the free tier of a retrieval platform makes the free tier usable by a whole team. |
| Krisp | Jul 2026 | Opened the first self-serve door in an otherwise application-gated developer line: the Voice Translation API moved from a 'Soon' badge inside the Apply-Now-only Voice AI SDK to a self-serve card with 61 languages, a 60-minute free translation credit, Python and JavaScript SDKs, a playground, a 99.9% uptime SLA and a 'Get API Key' button — no sales call. No per-minute rate is published beyond the free credit, so the free tier IS the published price. |
| ZenRows | Aug 2026 | The purest reversal of the app-layer free-tier deletions, inside the scraping cohort. ZenRows replaced its 14-day, $1-of-usage Free trial with a PERMANENT $0/month plan carrying 5,000 credits, as part of rebuilding the whole rate card around a shared credits currency across four primitives (Fetch 1x, JS rendering 5x, Premium Proxies 10x, both 25x). The paid ladder was rebuilt at the same time — Build $19-$39, Launch $69-$129, Growth $199-$399, Scale $549-$999, three rungs each, replacing Developer $69 / Startup $129 / Business $299 — and Residential Proxies was demoted to '(Legacy)'. Free went from a clock to a plan while the paid tiers got more complex. |
| Diffbot | Aug 2026 | A whole new product added to the free tier at no price. Diffbot launched a Web Search API as its fifth credit-metered product at 1 credit per query — the same rate as a page extraction — and bundled it into the existing 'all four APIs included' entitlement on EVERY plan, Free included, alongside Extract, Natural Language and Knowledge Graph. Headline prices held at Free $0 / Startup $299 / Plus $899 / custom Enterprise, credit allotments and overage rates were unchanged, and Crawl remains the sole feature gated to Plus and above. The relaunch also disclosed a previously undocumented mechanic: the Free plan does not auto-bill overage — exceeding 10,000 credits/month returns a 429 until the quota resets — where paid plans bill pro rata. |
| Frase | Aug 2026 | Free-tier creation on an adjacent content-API surface: FraseCMS, previously an unpriced bundled hosting perk with no disclosed caps, became a three-tier product with a genuine Free rung — $0 for 100,000 page views/month and 25 GB bandwidth on a Frase-badged subdomain — under Basic $19/mo (250,000 views, custom domain) and Pro $99/mo (1,000,000 views, 500 GB). Core plan prices held at Starter $49 / Professional $129 / Scale $299 monthly. A capped free entry published where none existed. |
Counterexamples
- Speechmatics · Aug 2026 — The cohort boundary, case 4 — and the second speech API to run Gladia's exact conversion. The Free and Pro plan cards no longer mention the recurring '3,000 minutes (50 hours) per month' speech-to-text allowance or the '1 million characters per month' of text-to-speech; both now read '$100 in credit to get started' (Pro: 'add a card when you're ready'), a one-time grant with no monthly reset. The page's 'Last updated' badge moved from 29 July to 31 July 2026. Not one rate moved — every per-hour STT SKU ($0.129-$0.43/hr), the $0.011/1k-character TTS rate, the bolt-on prices and the 20%/33% discounts are all unchanged. The entire change is the free tier, at a vendor whose evaluation traffic costs real audio-processing time.
- SambaNova · Aug 2026 — The cohort boundary, case 5, and the hardest form yet — the grant was deleted rather than converted. Every capture back through 2026-08-11 showed a Free tier reading 'Start exploring SambaNova Cloud with free API credits. No credit card required,' including '$5 in free API credits' and a page-wide banner promising '$5 of Free Credit... Initial credits will expire in 30 days.' As of 2026-08-14 that copy, the $5 wallet indicator and the expiry banner are all gone; the Free card now reads 'Start exploring SambaNova Cloud. Add a payment method and purchase credits to run your first requests.' Cerebras and Gladia converted recurring free access into a one-time credit; SambaNova removed the no-card path entirely.
- Novita AI · Aug 2026 — The free shelf emptied twice in a month on the same GPU-backed platform. On 2026-08-11 three LLMs tagged Time Limited Free converted to paid rates — Macaron V1 Venti to $1.5/M in / $4.5/M out, Macaron V1 Tall to $0.45/$2.6, Ling 3.0 Flash to $0.06/$0.18 — with Ling 3.0 Tiny added at $0/M as the replacement free listing. Three days later, on 2026-08-14, Ling 3.0 Tiny was delisted outright from both novita.ai/en/pricing and the model catalog, leaving no free LLM on the rate card at all and the catalog easing 177 to 174 models. The shortest lifespan of any free listing tracked here, and unlike Tencent's Hy3 in July it was removed rather than repriced.
- Linkup · Jul 2026 — Went the other way on its new surface: the closed-beta Extract endpoint requires a $10 minimum account balance to submit a task (variable ~$2-$10/task) — a premium extraction product gated behind a paid balance rather than opened to a free tier.
- Bright Data · Jul 2026 — Same vendor, opposite direction on its premium line: Bright Insights eCommerce-intelligence list prices jumped 5-8x ($250/mo and $400/mo → $2,000/mo each) the same day the commodity scraping APIs gained a free tier. The intelligence/premium product is gated upward even as the request meters open.
- turbopuffer · Jul 2026 — Partial counter: cut its floor but still has NO free tier, and its Scale ($256/mo) and Enterprise (≥$4,096/mo) minimums are unchanged — the land-grab is limited to the entry plan, not a free-forever tier.
- Cerebras · Jul 2026 — The cohort boundary, case 1: an inference API converting a recurring free tier into a one-time grant. The entry card was labelled simply 'Free' and pitched as 'the easiest way to get started with Cerebras' — open, rate-limited access across all models; it now reads 'Free Trial — get started with $5 in free credits after making an account'. A standing allowance became a single credit drop, removing the perpetual free path. Retrieval requests are near-free to serve; wafer-scale inference seconds are not.
- Gladia · Jul 2026 — The cohort boundary, case 2: a speech API retracting a recurring free tier. The 'Starter offers 10 hours of transcription free every month' allowance is gone, replaced by a prepaid credit wallet with auto-top-up opening on a one-time €50 signup credit with no monthly reset (Gladia puts it at 80+ async or 60+ real-time hours). Per-hour rates unchanged — async $0.61/hr, real-time $0.75/hr on Starter; as low as $0.20/$0.25 on Growth. The recurring-vs-one-time distinction this trend asks buyers to read flipped to the wrong side.
- Novita AI · Jul 2026 — The cohort boundary, case 3: a free MODEL going paid. Tencent's Hy3 was listed at $0/M input and $0/M output on 2026-07-06 and now bills $0.14/M input ($0.035/M cache read) and $0.58/M output, in the same sweep that pruned the catalog from 221 to 196 SKUs. On a GPU-backed serverless platform even a single free SKU is a real cost line, which is the inverse of the retrieval-API economics driving this trend.
- Gumloop · Jul 2026 — App-layer retraction at its sharpest: the permanent $0/month, 5,000-credit, 1-seat Free plan that let solo users try the product indefinitely was removed, replaced by a 14-day Pro trial available ONLY on monthly billing (switching to annual removes the trial CTA and goes straight to a paid signup). The Pro credit slider itself ($37 for 20,000 credits up to $1,840 for 1M) was unchanged across four capture cycles — so the on-ramp closed while the price held.
- Playground AI · Jul 2026 — Free-tier entitlements cut on both axes: downloads fell from 10 to 2 per day and the PGv3 rolling-window generation cap from 10 to 5 images per 3 hours, while Pro ($15/mo, $12 annual) and Pro Plus ($45/$36) held. Playground's own Free plan card still advertises 'up to 10 images every 3 hours' — an unresolved discrepancy on the live page. Creative-app free tiers are moving the opposite direction to retrieval APIs.
Trivia
-
Exa raised its free tier 20-fold in a single capture — from 1,000 to 20,000 requests per month (2026-07-14) — without touching its core per-1k-request endpoint rates, the largest single free-tier expansion in the retrieval/data-API cohort this window.
-
Unstructured converted a pure linear per-page meter into a capped one on both ends the same day: the 15,000 free pages now reset MONTHLY (was a one-time allowance) and Pay-As-You-Go bills are capped, above which pages are free up to 1M/month — so a 1M-page month bills at an effective floor near $0.003/page instead of the $0.03 list rate.
-
Bright Data ran the land-grab and the gate-up simultaneously: on 2026-07-14 it added a recurring 5,000-results/mo no-card free tier to Web Unlocker / SERP / Web Scraper, while on the same day raising its Bright Insights eCommerce-intelligence list floor 5-8x ($250 and $400/mo → $2,000/mo each) — commodity APIs opened, premium intelligence gated.
-
The free-tier direction split cleanly by what a free request actually costs the vendor. In the 2026-07-15 → 2026-07-30 window, 7 retrieval/dev-API vendors WIDENED (You.com, Firecrawl, Exa, Socket, Apify, LlamaIndex, Krisp) while the 3 API vendors that RETRACTED were all GPU- or audio-backed — Cerebras (Free tier → one-time $5 credit), Gladia (10 recurring free hours/month → one-time €50), Novita (Tencent Hy3 off $0/$0 to $0.14/$0.58). A search request is nearly free to serve; a GPU second is not.
-
Firecrawl (2026-07-30) added the corpus's flattest entitlement: an identical +1,000 search credits per month on every plan card from Free to custom Enterprise, printed as its own line under each tier's main credit pool. An Enterprise buyer paying an unpublished custom price gets exactly the same search-credit grant as a $0 signup — a deliberate refusal to scale one entitlement with price, consistent with Firecrawl's no-seats philosophy.
-
You.com moved its premium gate DOWN and its free floor UP inside eight days, which is the barbell running backwards: on 2026-07-22 it replaced the open-ended ">$2,000.00 /1k calls" label on its Contact-Sales Frontier research tier with a firm $1,200.00 (a ~40% cut plus a transparency gain), then on 2026-07-29 put a 100-calls/day free allowance in front of the flat $5.00/1k Web Search API. Its own docs still describe the old flat rate.
-
Exa widened its free tier twice in 15 days and changed its unit in the process: 1,000 → 20,000 requests/month on 2026-07-14, then on 2026-07-29 it deleted the requests-per-month framing entirely for dollar credits ($20 at signup, $10/month, no card required). The most aggressive free-tier expander in the cohort ended up denominating generosity in dollars rather than calls.
For buyers
For a usage-metered data or retrieval API, the free tier is now a real evaluation budget, not a token gesture — Exa's 20,000 requests/month and Unstructured's recurring 15,000 pages can run a modest production workload at no cost. Read the free tier on two axes: recurring-vs-one-time (a monthly reset like Unstructured's or a daily pool like Powerdrill's beats a one-week trial) and no-card-vs-trial (Bright Data's 5,000/mo needs no credit card). Check whether the bill is capped at scale (Unstructured caps PAYG, above which pages are free to 1M/month) or whether the floor is a hard minimum (turbopuffer's $16 entry plan, Linkup's $10 Extract balance). Then budget the premium/intelligence tier as a separate line — that is exactly where these vendors are raising prices even as the commodity meters open, so the cheap eval does not predict the cost of the high-value surface.
For vendors
Running this play needs a metering layer that can carry a genuinely recurring free allowance — a monthly reset, a daily-refreshed credit pool, or a no-card monthly grant — without leaking into production margin, since the whole logic rests on the marginal cost of an eval workload being near zero. The reusable move is to widen the free/eval entry while holding your per-unit rates flat (Exa 20×'d the free tier but did not touch endpoint rates), so acquisition improves without repricing revenue. The barbell is deliberate: open the commodity request meters to a land-grab and gate the premium/intelligence surface upward in the same release (Bright Data did both on 2026-07-14), so the free tier is customer acquisition and the intelligence product is where willingness-to-pay is captured. The tradeoff is durability — a generous free tier is easy to match, so if the widening is only a response to crowded search/scraping competition it becomes table stakes rather than an edge, and the margin has to come from the gated tier and from converting eval volume to production.
Outlook — what to watch
Logged new in July 2026 on a single 2026-07-14 change cluster — five vendors widening the free/eval entry against three counter-current moves — so it is a same-week signal, not yet a corpus-wide law. It sharpens into a confirmed pattern if the next change batches show more retrieval, scraping, extraction, and vector-DB vendors enlarging recurring free tiers or cutting entry floors while holding per-unit rates, and if the barbell persists — commodity meters opening while intelligence surfaces gate up. It weakens if the free-tier widening proves to be a one-window response to crowded competition and vendors quietly claw the allowances back, or if the land-grab spreads to the premium surfaces too (dissolving the barbell). The open question is whether generous recurring free tiers are a durable acquisition strategy for data/retrieval infrastructure or a temporary truce in a price war; the thing to watch is whether the free entries stay wide once the competitive pressure eases.
Bottom line
In one 2026-07-14 window, five retrieval/data-API vendors widened the free/evaluation entry — Bright Data added a recurring 5,000-results/month no-card free tier, Exa 20×'d its free tier to 20,000 requests/month, Powerdrill put a 1,000 daily-credit pool on every tier, Unstructured made its 15,000 free pages a monthly reset (and capped PAYG bills), and turbopuffer cut its entry floor 75% to $16 — all without touching core per-unit rates. The same vendors gate their premium intelligence surfaces upward (Bright Insights +5–8×, Linkup Extract behind a $10 balance), so the shape is a barbell: commodity request meters open to a land-grab, high-value intelligence gates up.
FAQ
Are data and retrieval API vendors expanding their free tiers in 2026?
Yes — in the 2026-07-14 change window alone, five did at once. Bright Data added a recurring 5,000-results/month no-card free tier to its scraping APIs, Exa raised its free tier 20-fold (1,000 → 20,000 requests/month), Powerdrill put a 1,000 daily-refreshed-credit pool on every tier including Free, Unstructured turned its 15,000 free document pages into a monthly reset, and turbopuffer cut its cheapest paid floor 75% ($64 → $16). None of them cut their core per-unit rates — the competition is on the width of the free/evaluation entry, not the marginal price.
Why widen a free tier instead of cutting per-unit prices?
For usage-metered data and retrieval infrastructure the marginal cost of a small evaluation workload is near zero, so a generous recurring free tier is cheap customer acquisition — the revenue comes at production volume. Widening the free entry lands developers to try and build on the API without giving up pricing power on paid usage, which is why vendors like Exa expanded the free tier 20× while leaving endpoint rates untouched.
What is the 'barbell' in data-API free-tier pricing?
The barbell is that the same vendors open their commodity request meters to a free land-grab while gating their premium and intelligence surfaces upward. On 2026-07-14 Bright Data added a free tier to its scraping APIs and, the same day, raised its Bright Insights eCommerce-intelligence list 5–8× ($250 and $400/mo → $2,000/mo each); Linkup opened nothing free and put its new Extract endpoint behind a $10 minimum balance. Commodity meters open; high-value intelligence gates up.
How should I read a data-API free tier before I commit?
Check three things. First, recurring-vs-one-time: a monthly reset (Unstructured's 15,000 pages) or a daily-refreshed pool (Powerdrill's 1,000 credits) can run a modest workload indefinitely, unlike a one-week trial. Second, no-card-vs-trial: Bright Data's 5,000/mo needs no credit card. Third, the scale behavior: is the bill capped (Unstructured caps PAYG, above which pages are free to 1M/month) or is there a hard floor (turbopuffer $16, Linkup $10 balance)? Budget the premium/intelligence tier separately — that is where prices are rising even as the free meters open.