Free tiers moved in both directions at once — and layer only half explains it
Free-tier moves are running in opposite directions simultaneously: of 23 dated moves, 16 narrowed a free tier and 7 widened one, and 22 of the 23 landed inside a single 31-day window. The binding constraint is measurement — the corpus field has_free_tier cannot see any of it, because a permanent recurring free tier and a 14-day trial both set it true.
What's happening — and why
What's happening: the free tier stopped being a fixed acquisition constant and became an active monetization lever that vendors adjust in both directions, often within days of each other.
The original reading was that direction tracks what the vendor sells — seat-shaped applications narrow, developer-facing APIs widen — and 13 of the 16 narrowing moves were applications while all 7 widening moves were APIs or dev tools. But that tilt is our own coding of documented moves rather than a vendor-stated rationale, and the 2026-08-17 window ran the exact inverse on four dated moves: the two largest widenings were seat-shaped applications (Semrush adding a permanent $0/mo plan, Regie.ai relaunching as self-serve freemium) while two developer APIs narrowed (MiniMax closing its Music and Lyrics APIs to new signups, Apify cutting free concurrency 25 to 5).
So the layer heuristic should not be used predictively. What survives is the measurement problem, and it is the durable finding: the corpus tracks a boolean that collapses a permanent recurring allowance and a two-week trial into the same value.
How it works
Evidence over time
34 supporting · 5 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Poe | Mar 2026 | NARROWED (app/seat layer, and the precedent for the wave that followed). The Free tier's daily compute-points allowance fell from ~3,000/day to ~300/day — a ~90% cut — with no changelog entry, help-centre update or announcement; the date and magnitude are corroborated by three independent third parties, not by Poe. The same lever was pulled on the paid side on 2026-07-28: Premium's monthly allowance went from 1,000,000 to 660,000 compute points (-34%) with the $19.99 sticker price unchanged. |
| Lokalise | Jun 2026 | NARROWED (app/seat layer). The perpetual $0 plan was retired in the same release that CUT Growth from $499 to $375/mo — price down, door closed. Explorer at $144/mo becomes the entry point and a 14-day no-credit-card trial is the only free path in. Growth's bundled allowances shrank too: 300K to 225K processed words/yr and 50K to 40K Pro AI words/yr. |
| Qodo | Jun 2026 | NARROWED (app/seat layer). The permanent free Developer tier was dropped alongside per-seat pricing ($30/user/mo annual), replaced by a 14-day trial and a team-wide pooled credit balance at $0.012/credit sold in 2,500 / 5,000 / 20,000-credit packs. Qodo now states it does not offer a free plan. |
| Speechmatics | Jul 2026 | WIDENED (developer API layer), and the direct counterpart to Gladia 16 days later. The free monthly allowance grew from 2,400 to 3,000 minutes (50 hours: 1,200 real-time + 1,800 batch) plus 1M free TTS characters, no card required, while real-time enhanced speech-to-text was cut from $0.56 to $0.43/hr and a new multilingual Batch Melia 1 model landed at $0.129/hr. |
| Bright Data | Jul 2026 | WIDENED (developer API layer) — and it is the exact inverse of the app-layer move. Web Unlocker, SERP API and Web Scraper API each gained a genuinely recurring 5,000-results-per- month free tier with NO credit card, replacing a one-week, 1,000-record one-time trial. The old $1.1 and $1.0 per-1k committed rates were removed in the same restructure ($1.5/1k PAYG, a single $499/mo Scale tier at ~380K included with $1.3/1k overage, Enterprise by quote). |
| Unstructured | Jul 2026 | WIDENED (developer API layer), and in precisely the dimension Cerebras and Gladia narrowed: the 15,000 free pages became a MONTHLY reset instead of a one-time allowance. A $3,000/mo bill cap was added on top, above which pages are free up to 1M pages/mo, against a flat $0.03/page list rate. |
| Artisan | Jul 2026 | NARROWED (app/seat layer) — the hardest single retraction in the window. The public self-serve credit rate card published earlier in 2026 (Free $0/300 credits, Intern $250/mo /12K credits, Employee $600/mo/30K credits, plus a credit estimator and a 10,000-credit $300 free trial) was replaced by three sales-gated lead-volume tiers — Team ~2,500 leads/mo, Scale ~6,000, Enterprise — every card reading "Talk to sales." No Free tier, no estimator, no self-serve checkout. Its second retreat from public pricing in two years. |
| 6sense | Jul 2026 | NARROWED (app/seat layer) — and the fastest reversal in the window. The free Sales Intelligence tier (50 data credits/mo, company & people search, sales alerts, list builder, Chrome extension, self-serve "Start for Free") was removed eight days after the 2026-07-06 capture that documented it. The page now shows only three quote-based bundles, all routing to Book a Demo, with no published dollar figures and no self-serve entry. |
| Freshworks | Jul 2026 | NARROWED (app/seat layer). The $0 three-user Freshsales Free plan is gone from the page, ending a freemium ladder Freshsales had run since its 2022 post-IPO repackaging; only a 21-day no-credit-card trial remains, and the FAQ states the account is suspended if no plan is chosen at trial end. Seat prices did not move ($9/$39/$59 per user/mo annual) — and the same release printed add-on prices for the first time (Freddy AI Agent $49 per 100 bot sessions, Branded documents/CPQ $19/user/mo). |
| Puzzle | Jul 2026 | NARROWED (app/seat layer), via a single footnote. "Starter plan is free until $20k of transaction volume" — a genuine usage threshold that let a pre-revenue company stay on real books indefinitely — became "Starter is free for 2 months," paired with a struck-through $25 list price and a "100% off for 2 months" badge. Starter and Core AI credits were relabelled "25 credits life time" where they previously read "25 credits". Subscription prices unchanged. |
| Cognition | Jul 2026 | NARROWED (app/seat layer) without deleting the tier. Devin's Free plan CTA became "Download" (Devin Desktop) instead of a sign-up, and every Devin Cloud row — cloud agents, DeepWiki, Devin API, Ask Devin, concurrent sessions — is marked not-included on Free. Headline prices held at $0/$20/$200 and $80 + $40 per full dev seat: the entitlement moved, the price did not. |
| Cerebras | Jul 2026 | NARROWED — and it is a LAYER CROSSER, a usage-metered inference API that time-boxed free anyway. The card was relabelled from "Free" ("the easiest way to get started with Cerebras," unmetered across all models subject to rate limits) to "Free Trial": $5 in credits after account creation. The docs add two conditions the marketing card omits — the credits arrive only after adding a verified payment method, and they expire 30 days after they are granted. A standing perpetual allowance became a one-time, carded, expiring grant. |
| Novita AI | Jul 2026 | NARROWED — the second LAYER CROSSER, and a different shape: free is a property of a SKU, not of an account. Tencent's Hy3 moved off $0/$0 to $0.14/M input and $0.58/M output, emptying Novita's free-LLM shelf entirely (only an audio model, MOSS TTS v1.5, remains a Free listing). The catalog was pruned 221 to 196 models in the same capture. No account changes state, so nothing in `has_free_tier` registers the loss. |
| Gladia | Jul 2026 | NARROWED — the third LAYER CROSSER, and the cleanest recurring-to-one-time conversion in the corpus. "10 hours of transcription free every month" was replaced by a single non-resetting 50 EUR signup credit (Gladia puts it at ~80+ async or 60+ real-time hours), alongside a move from arrears billing to a prepaid wallet with auto top-up. Per-hour rates did not move: $0.61 async / $0.75 real-time on Starter. |
| Photoroom | Jul 2026 | NARROWED (app/seat layer). The Free plan card was dropped from the app pricing page (the subhead still reads "Start for free") and the Weekly billing option disappeared, in the same capture that replaced relative multipliers with absolute allowances (8,000/25,000/75,000 AI credits and 1,000/3,000/10,000 exports per month). Plan dollar fields still render blank, and the credit FAQ now warns the monthly credit amount "may vary depending on your country or region." |
| Maxio | Jul 2026 | NARROWED (app layer, developer-facing surface). The free "Build / Try for free" developer- sandbox column was removed from the three-plan compare table and its standalone callout deleted; the sandbox survives only as a footer link, not a listed plan. Grow held at $599/mo (up to $100k monthly billings), leaving no free option merchandised on the pricing page. |
| Exa | Jul 2026 | WIDENED (developer API layer), for the second time in a month. The signup credit doubled $10 to $20, the recurring monthly credit rose $7 to $10, and the payment-method requirement for the monthly credit was dropped — so every Free Tier account now gets it, not only carded ones. The "20,000 requests/month" framing was replaced by pure credits. It had already expanded on 2026-07-14 (free tier 1,000 to 20,000 requests/mo, a 20x move). |
| You.com | Jul 2026 | WIDENED (developer API layer) by adding a free tier where none existed. The Web Search API card was restructured from a flat $5.00 per 1,000 calls to "Free / 100 calls per day," with the unchanged $5.00/1k rate applying only above that daily quota. The API docs had not yet been updated to mention the free quota at capture time. |
| Socket | Jul 2026 | WIDENED (developer tool) — and it is the crosser on the widening side, because Socket is a per-developer subscription rather than a usage meter. The Free plan's tracked-dependency cap rose 1,000 to 10,000 (10x) between the June and July 2026 archived snapshots, while Team ($25/developer/mo) and Business ($50/developer/mo) list prices held steady. |
| Hyperline | Jul 2026 | NARROWED (app layer). "10 invoices free, no credit card" is gone and the new Launch/Growth/Scale ladder has no free tier at all. The entry price moved from $199/mo + 0.6% of billed revenue to a flat $599/mo (up to $2M annual revenue, 100 active contracts), with Growth and Scale custom-quoted — a take-rate abandoned and a free door closed together. |
| Playground | Jul 2026 | NARROWED (app/seat layer) by squeezing rather than deleting. Free downloads fell 10 to 2 per day and the rolling PGv3 generation cap fell 10 to 5 images per 3-hour window; the promo-code field was removed from the Pro checkout cards. Pro ($15/mo) and Pro Plus ($45/mo) are unchanged. Free's own plan-card copy still advertises "up to 10 images every 3 hours" — an unresolved live discrepancy on Playground's page. |
| Gumloop | Jul 2026 | NARROWED (app/seat layer), and the purest on-ramp deletion in the window. The permanent $0/mo, 5,000-credit, 1-seat Free plan was removed and replaced by a 14-day Pro trial offered ONLY on monthly billing — selecting Annually removes the trial CTA and goes straight to a paid signup button. The Pro slider ($37/20k credits up to $1,840/1M credits) was confirmed unchanged across four capture cycles (2026-06-02, 06-30, 07-23, 07-30), so nothing about the price moved: only the door. |
| Firecrawl | Jul 2026 | WIDENED (developer API layer) on the same day Gumloop narrowed. Every plan card — Free, Hobby, Standard, Growth, Scale and custom Enterprise — gained a flat "+1,000 search credits" line beneath its main allotment, identical regardless of plan size, on top of the unchanged 2-credits-per-10-results Search rate. Headline plan fees ($0/$16/$83/$333/$599/Custom) were re-confirmed unchanged. |
| Speechmatics | Aug 2026 | NARROWED — and this is the same company listed above as a WIDENING case on 2026-07-06, which makes it the trend's own evidence table reversing inside four weeks. The Free and Pro cards no longer mention the recurring "3,000 minutes (50 hours) per month" STT allowance or the "1 million characters per month" of TTS; both now read "$100 in credit to get started," a one-time grant with no monthly reset, with Pro adding "add a card when you're ready." The page's Last-updated badge moved 29 July to 31 July 2026. No rate moved: every per-hour STT SKU ($0.129-$0.43/hr), the $0.011/1k-character TTS rate, the bolt-ons and the 20%/33% discounts are identical. It is Gladia's conversion, run by the vendor this trend had cited as Gladia's opposite. |
| SambaNova | Aug 2026 | NARROWED — a fifth layer crosser, and the hardest form: deletion rather than conversion. 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." All three are gone; the Free card now reads "Start exploring SambaNova Cloud. Add a payment method and purchase credits to run your first requests," and its feature list has shrunk to pay-as-you-go credits. Cerebras and Gladia converted recurring free access into a one-time grant; SambaNova removed the no-card path entirely at a custom-silicon inference vendor. |
| Mintlify | Aug 2026 | NARROWED by unbundling the capability rather than the tier. Mintlify kept a genuinely free Starter plan with the full documentation platform (custom domain, web editor, MCP server, API playground) but moved the AI assistant, writing agent, agent skills and automations off it entirely, and Starter's 5,000-credit monthly allotment no longer appears in the feature table at all. It is now capped at 5 editor seats. The features landed in a reintroduced Pro tier at $450/mo billed annually or $540/mo monthly with 10,000 AI credits at the same $0.01/credit overage. `has_free_tier` stays true; the free tier lost the entire AI product. |
| ZenRows | Aug 2026 | WIDENED (developer API layer), and it is Gumloop's 2026-07-30 deletion run exactly backwards. ZenRows replaced its 14-day, $1-of-usage Free trial with a PERMANENT $0/month plan carrying 5,000 credits, inside a full rebuild from dollar-denominated CPM / per-GB / per-session-hour pricing to a single credits currency spanning 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 — and Residential Proxies was demoted to "(Legacy)". The door opened while the ladder got more complex. |
| WellSaid | Aug 2026 | WIDENED by de-timing a clock — the inverse of the five permanent-plan-to-trial conversions logged above. The free Trial card dropped its "try all our voices for 7 days" copy entirely and now states "3 download minutes per month (no commercial rights)," and the comparison table's downloaded-minutes row moved from "No downloads" to "3 minutes/month" for the Trial column. A converting, time-boxed trial became an evergreen free tier with a small recurring allowance. Starter and Pro each gained an "Extra minutes available" line — WellSaid's first published self-serve overage — under a sitewide "New plans are here" banner. |
| Nomic | Aug 2026 | WIDENED at the hardest possible starting point: a flagship product with no self-serve path at all. Nomic's AEC Platform had sold only Business ($40/user/month, 25-seat minimum, an explicit $1,000/month platform commitment) and custom Enterprise since its November 2025 repositioning. The page now lists four tiers, adding Free (no credit card required, limited agent requests, core Platform surfaces) and Individual ($20/month, self-serve, everything in Free plus extended agent requests and all product surfaces except the Agent API, with AI usage funded by on-demand top-ups). The standalone $1,000/mo platform-commitment line was dropped from the page. |
| Wispr Flow | Aug 2026 | Both directions on one offer in eleven days, which is the trend's core observation compressed into a single vendor. Four new FAQ entries appeared in the Flow Discounts help-centre article around 2026-07-30 (captured 2026-08-04) describing a verified-student-only FREE, unlimited-words annual plan claimed via a "Claim free plan — no credit card required" card in the desktop-app sidebar, and the Who-qualifies bullet changed to "Free unlimited words on an annual plan." By the 2026-08-11 capture the article had reverted to "Student: Discounted Pro pricing (availability may vary — contact support to check eligibility)" with its original two-question FAQ set; the free card and all four FAQ entries are gone. |
| Semrush | Aug 2026 | A direct counter-current to this trend's central tilt: a seat-shaped, subscription-priced application WIDENING rather than narrowing. Semrush added a permanent $0/mo Free plan (1 demo project, 10 reports/day across all vital tools, Help Center and Basic Onboarding, no credit card, AI Visibility features excluded) in place of trial-only free access on its four paid SEO + AI Search tiers. Its corpus has_free_tier flipped false to true. |
| Regie.ai | Aug 2026 | The largest single widening in the corpus this window, and again at a seat-shaped application. Regie.ai relaunched as RegieGO, replacing sales-led annual seats ($180/user/mo AI SEP with a 10-seat minimum; $499/user/mo Force Multiplier Rep with a 5-seat minimum; no free tier; every purchase through "Talk to sales") with a self-serve Free plan ($0, 250 one-time credits, no card) plus Pro at $49/mo ($41 annual, 5,000 credits/mo) and a custom Enterprise. A WELCOME10K promo adds 10,000 credits. has_free_tier false to true and has_commits true to false in the same change. |
| MiniMax | Aug 2026 | Narrowing on the API side, which fits the tilt: MiniMax closed its paid Music and Lyrics Generation APIs to new signups effective 2026-08-20, grandfathering existing paying customers, and retired the free music-generation tiers outright. A developer-facing API removing free access entirely, in the same pass that raised Token Plan subscriptions 10%. |
| Apify | Aug 2026 | Narrowing on the API side by capacity rather than by price: Free-plan concurrent Actor runs cut 25 to 5 (-80%) with every paid tier's concurrency, every subscription price and every compute/proxy/storage rate unchanged. |
Counterexamples
- Cerebras · Jul 2026 — Listed in evidence above as a narrowing move, and repeated here deliberately: it is evidence FOR the two-directions observation and a counterexample AGAINST the layer explanation. A usage-metered inference API with no seats converted an open, rate-limited Free tier into a one-time $5 credit trial — the move the layer story predicts only for seat-shaped apps. Novita (2026-07-21) and Gladia (2026-07-22) are the other two crossers.
- Diffbot · Apr 2024 — Ran Gumloop's 2026-07-30 move exactly backwards, two years earlier: a permanent free-forever plan (10,000 credits/mo at $0/credit, no credit card) REPLACED the 14-day trial, and the page headline changed from "Plans starting at $299/mo" to "Plans starting at Free." Proof the direction is reversible and not a one-way ratchet.
- GitHub Copilot · Dec 2024 — A $0 Free plan was LAUNCHED (and Copilot Individual renamed Pro at the unchanged $10/mo) — a seat-shaped app adding a free tier, which the layer explanation would not predict. Copilot still does not publish the Free tier's credit allowance ("limited chat and agent usage"), so even where a free tier is added the entitlement can stay unobservable.
- Gorgias · Jul 2026 — A third shape neither arm captures: Gorgias has no free tier at all (`has_free_tier` false), yet it bundled an AI Agent allotment into every paid plan (30/30/190/530 automated interactions) with overage at a flat $1.50 per automated interaction, up from $0.90. The AI capability became free-at-the-margin inside paid tiers — free access moving without a free tier existing.
- Helicone · Jan 2025 — The mid-arc case that neither "killed" nor "widened" describes: the free monthly request allowance was cut from 100,000 to 10,000 (a 90% reduction) and the free tier then simply stayed there — Hobby is still free forever at 10,000 requests/month in 2026. Narrowing does not have to end in deletion.
Trivia
-
The `has_free_tier` field held at exactly 215 companies while its rate fell from 64% (215 of 338) to 61% (215 of 353) — the 15 companies added in between contributed zero free tiers, so the percentage moved three points with no vendor changing anything. In the same stretch five vendors converted a permanent free plan into a clock (Lokalise 14-day on 2026-06-30, Qodo 14-day on 2026-06-30, Cerebras $5 credit trial on 2026-07-21, Puzzle two-month promo on 2026-07-21, Gumloop 14-day on 2026-07-30) and the boolean did not move for any of them.
-
2026-07-30 is the corpus's sharpest single-day contradiction on this axis: Gumloop deleted a permanent $0/mo, 5,000-credit, 1-seat Free plan in favour of a 14-day Pro trial offered ONLY on monthly billing (switching to annual removes the trial CTA entirely), while Firecrawl added a flat +1,000 search credits per month to every plan card — Free through custom Enterprise, identical regardless of plan size. Neither vendor changed a headline price on the day.
-
Two speech-to-text APIs moved in opposite directions 16 days apart, which the layer explanation cannot separate: Gladia (2026-07-22) replaced a recurring 10-free-hours-per-month allowance with a single non-resetting 50 EUR signup credit, while Speechmatics (2026-07-06) grew its recurring free allowance from 2,400 to 3,000 minutes a month and cut real-time enhanced from $0.56 to $0.43/hr. Same layer, same modality, opposite direction, same month.
For buyers
Never treat 'has a free tier' as a fact about a vendor; treat it as a question with four parts. Is the allowance recurring or one-time? Is it capped by volume, by concurrency, or by time? Does it require a card? And can it be withdrawn without a price change? Those four answers move independently and none of them appears in a comparison table. Assume free-tier terms are no more durable than a mid-tier price — in one 15-day window, free-tier terms moved 12 times across the corpus.
For vendors
The free tier is the cheapest lever you have and the one most likely to be read as bad faith, because buyers experience it as a promise rather than a plan. If you must narrow, narrow on a dimension the buyer can plan around — a volume cap is easier to absorb than a concurrency cap, and a concurrency cap is easier than converting a recurring grant to a one-time one. If you widen, note that the corpus contains several vendors who widened substantially and got no credit for it, because the change was invisible to every tracker watching prices.
Outlook — what to watch
Expect continued two-way movement and no convergence, because the free tier is now doing three different jobs across the corpus — evaluation access for developer APIs, top-of-funnel acquisition for applications, and competitive positioning for incumbents adding AI. The trend's layer heuristic is already partially falsified and should be re-tested rather than restated. The measurement gap is the thing most likely to change: a corpus field that distinguished a recurring allowance from a trial would make this trend properly countable for the first time.
Bottom line
Free tiers moved 16 times narrower and 7 times wider in a single window, and the corpus records every one of those 23 moves as the same boolean. The layer heuristic that seat-shaped apps narrow and APIs widen was inverted by four dated moves in the latest window and should not be used predictively.
FAQ
Are AI free tiers getting more or less generous?
Both, at the same time. Of 23 dated moves, 16 narrowed a free tier and 7 widened one, with 22 of the 23 landing inside a single 31-day window. There is no net direction to quote — the useful reading is that free-tier terms are now repriced as actively as paid tiers.
Do developer APIs have more generous free tiers than apps?
That was the founding tilt — 13 of 16 narrowing moves were seat-shaped applications and all 7 widening moves were developer APIs — but the latest window ran the inverse on four dated moves. Semrush and Regie.ai, both seat-shaped applications, added permanent free tiers, while MiniMax and Apify, both developer-facing, narrowed. Use the heuristic descriptively, not predictively.
Why can't the corpus measure this directly?
Because has_free_tier is a boolean, and a permanent recurring free tier and a 14-day trial both set it to true. A vendor can convert one into the other — as Gladia did, swapping a recurring 10-hours-per-month allowance for a one-time 50 EUR credit — with no change to the field at all.
What should I check before relying on a free tier?
Whether the allowance recurs or is granted once, whether the cap is on volume, concurrency or time, whether a card is required, and whether the vendor has changed those terms before. Unstructured converted 15,000 free pages every month into 10,000 pages once at signup while halving its paid per-page rate — a change that made evaluation materially harder and production materially cheaper, in one move.