From subscriptions to usage — and back to hybrid, sometimes within six weeks
Several vendors abandoned flat subscriptions for pure pay-as-you-go -- then re-introduced tier structure on top. The stable destination is hybrid (a seat or platform fee plus metered overage), where 144 of 338 corpus companies (43%) sit -- the single largest paid structure -- while a freemium overlay (169) is the most common arrangement overall.
What's happening — and why
What's happening: several vendors dropped flat subscriptions for pure pay-as-you-go — then quietly added tiers and base fees back on top. The common landing spot is 'hybrid': a platform or seat fee plus metered overage.
Why: pure usage aligns price with value but makes revenue (for the vendor) and spend (for the buyer) hard to predict. A base fee restores that predictability while metering still captures heavy users — so the market converges on a blend rather than either extreme.
How it works
Evidence over time
17 supporting · 10 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Exa | Jul 2024 | Dropped subscription tiers entirely for pure pay-as-you-go credits. |
| Tavus | Aug 2024 | Pivoted from a flat $275/mo 'Intro' plan to usage-based CVI API pricing. |
| ElevenLabs | May 2025 | Began an explicit shift toward pay-as-you-go metering. |
| Deepgram | Oct 2022 | Restructured to prepaid PAYG / Starter / Growth credit packs. |
| Lindy | Mar 2026 | Pivoted from credit-pack PAYG model to a flat subscription ladder (Plus $49.99, Pro $99.99, Max $199.99) in March 2026 — subscription added back on top of the usage logic, illustrating the bidirectional nature of the drift. |
| Synthflow | Jun 2024 | Launched as bundled-minutes subscription tiers (Starter/Pro/Growth/Agency) after pivoting from earlier PAYG positioning; the per-minute meter is wrapped in a subscription structure rather than exposed directly. |
| Dust | Jun 2026 | Scrapped a ~2-year-stable flat €29/seat unlimited plan for credit-metered Free/Pro($30)/Max($150) seats — every message now consumes credits against a monthly allowance, with $0.01/credit programmatic overage. A clean subscription→usage move that kept the seat/plan envelope (hybrid), not pure PAYG. |
| Qodo | Jun 2026 | Replaced per-seat Teams pricing with a pooled credit balance ($.012/credit, packs of 2.5K/5K/20K) plus purchasable overage against a buyer-set cap — flat per-seat subscription drifting to pack-bundled usage metering, landing on the hybrid attractor. |
| Jasper | Jul 2026 | A three-year round-trip closed: "Jasper has reintroduced usage metering three years after killing the word caps it launched with." It was a pure per-seat subscription with no meter of any kind — Pro $59/seat/yr-billed ($69 monthly), Business quoted. Now Business carries a credit meter at 10 credits per Jasper Grid row, 10 per GEO Hub query or page run, 40 per Research or Translation Agent run, 100 per Optimization Agent run, and 1-40 per API/MCP call, with the seat price unchanged and NO dollar-per-credit rate published. The FAQ states credits "sit alongside seat-based pricing on the Business plan" — the hybrid attractor described in the vendor's own words. Source: changes/jasper-2026-07-22-packaging.md. |
| Linear | Jul 2026 | Incremental drift with the seat price frozen: 'Linear Agent automations (beta)', previously free inside the Business seat, was renamed 'Loops' and now carries the same "** Requires AI credits" footnote as Coding Sessions. Seat prices unchanged (Free $0, Basic $10/user/mo, Business $16/user/mo). This makes Loops the SECOND AI capability Linear meters outside the flat seat; Code Intelligence and Triage Intelligence stay bundled. The drift here is not a repackaging event but a slow migration of individual features across the bundled/metered line. Source: changes/linear-2026-07-29-packaging.md. |
| Vellum | Jul 2026 | The walk-back direction, inside three months: hosted pricing had been a bare configurator since Vellum's May 2026 pivot ($10/mo platform fee + machine tier + storage tier, effective $50/mo minimum, no named packages). It now carries three named presets that bundle fee, machine, storage and a monthly credit allowance into one price — Mighty $30/mo ($25 credits), Super $100/mo ($45 credits), Ultra $200/mo ($115 credits) — with the a-la-carte path demoted to 'Custom'. A configurator converging back into subscription packages with usage allowances inside them. Source: changes/vellum-2026-07-29-packaging.md. |
| Glean | Jul 2026 | Enterprise-scale version of the same drift: Glean expanded its Enterprise Flex rate card to 13 metered capabilities — a seat-and-platform enterprise search vendor publishing a multi-line consumption rate card beside the license rather than absorbing AI cost into it. Source: changes/glean-2026-07-22-packaging.md. |
| Composio | Jul 2026 | Six-dimension usage pricing announced effective 2026-08-15, with the $229 tier retired and a $599 Business tier added — the subscription ladder and the multi-axis meter being rebuilt in the same release rather than one replacing the other. Source: changes/composio-2026-07-23-packaging.md. |
| Chargebee | Aug 2026 | The attractor turned into a customer-facing toggle. Chargebee retired the Starter (USD 0/mo, free to USD 250K cumulative billing, then 0.75%) and Performance (USD 7,188/yr) tiers for a single plan, Flow, sold as a CHOICE between this trend's two poles: pay-as-you-go at 0.80% of monthly billing value with a $0 platform fee, or Commit monthly at a $99/mo platform fee plus 0.65%. Same product, same 100M usage-events/month allowance either way; the buyer picks whether to buy pure-usage or platform-fee-plus-discounted-usage. Enterprise Plus is quoted at 500M events. The USD 250,000 free threshold, in place since at least 2022, was removed, so the percentage applies from the first dollar. No other corpus vendor has published the hybrid decision as a pricing-page option rather than as a structure. |
| Lindy | Aug 2026 | Both halves of the drift in one revision. Lindy moved every tier from a flat per-account fee to PER-SEAT billing (Plus $29.99/user, down 40% from $49.99; Pro $99.99/user; Max $199.99/user) AND published its credit meter for the first time — 3,000 / 15,000 / 35,000 credits per tier, pooled across the whole team, with named credit bands by task size replacing an undefined 'standard usage' allowance. The struck-through '$8,000/month human assistant' anchor that had headlined the page was deleted. This is the fourth structural repricing since Lindy pivoted from a credit-billed agent builder to a flat-subscription assistant in March 2026, and it lands squarely on seat-plus-meter. |
| Framer | Aug 2026 | A website builder puts a meter beside its site plans. Framer introduced fixed monthly Agents AI-credit pools — 500 / 1,000 / 3,000 credits by tier — to meter its agentic AI site-generation features, and simultaneously dropped the intermediate Scale plan, leaving Free / Basic ($10/mo yearly) / Pro ($30/mo yearly) / Enterprise site plans alongside per-seat Workspace plans. It follows an Oct-Nov 2025 simplification (Mini tier removed) that drew community pushback: the ladder shrinks while a new metered axis appears next to it. |
| ZenRows | Aug 2026 | Meter and ladder rebuilt together, not one replacing the other. ZenRows replaced dollar-denominated CPM (Scraper API), per-GB (Residential Proxies) and per-session-hour (Scraping Browser) billing with a single shared CREDITS currency spanning four renamed primitives — Fetch, Extract (beta), Batch (beta) and Browser Sessions — at published fixed weights (1 credit standard, 5 with JavaScript rendering, 10 with Premium Proxies, 25 for both). It kept a full subscription ladder around it: Free ($0/mo, permanent, 5,000 credits — previously a 14-day trial), Build $19-$39, Launch $69-$129, Growth $199-$399 and Scale $549-$999, three rungs each, with Enterprise above 12.5M credits/month. Residential Proxies was demoted to '(Legacy)'. |
Counterexamples
- Bland AI · Dec 2025 — Re-coupled per-minute usage to subscription plan tiers (Build $299 / Scale $499).
- Exa · Apr 2026 — Re-introduced structured per-endpoint pricing cards atop the PAYG base.
- Runway · May 2026 — Stayed subscription-first — per-seat tiers with bundled monthly credits plus a usage API.
- Daily · Jun 2022 — Shifted to pure PAYG from subscriptions and stayed there — a clean drift to pure-usage without the walk-back to hybrid.
- Resemble AI · Jul 2026 — The corpus's fastest reversal, and the strongest single counterexample to a one-way reading of this trend: six weeks after collapsing a five-tier subscription ladder into a single never-expiring PAYG Flex plan, Resemble reintroduced Team ($280/mo annual / $350 monthly, 5 seats) and Business ($800/mo annual / $1,000 monthly, 20 seats, plus SSO) above Flex, both carrying discounted usage rates. The Flex detection rate also moved $0.04 → $0.035/sec and the sales nudge threshold dropped from $2,000/month of Flex spend to $1,000/month on Business. Pure PAYG lasted six weeks. Source: changes/resemble-ai-2026-07-29-packaging.md.
- Hyperline · Jul 2026 — RECAST 2026-08-16 — this was cited as 'a clean usage→subscription reversal' and it did not survive six days. On 2026-07-29 Hyperline abandoned the percentage-of-billed-revenue take rate it had used since its 2023 launch ($199/mo + 0.6% and $299/mo + 0.7%) for flat tiers gated on the CUSTOMER's revenue band and active-contract count — Launch $599/month flat, Growth and Scale custom-quoted — and dropped the 10-invoices-free trial. See the 2026-08-04 entry for the reversal. The move is retained as evidence that a vendor CAN replace a meter with a subscription, and as the setup for the fastest cross-line round trip in the corpus.
- Hyperline · Aug 2026 — THE FASTEST CROSS-LINE ROUND TRIP IN THE CORPUS: six days. Hyperline's flat $599/month Launch tier, introduced 2026-07-29 as this trend's cleanest usage→subscription case, is gone. Launch is now $299/month PLUS 0.7% of revenue processed — close to but not identical with the $199+0.6% / $299+0.7% structure that preceded the July change. The eligibility bands are unchanged: Launch still caps at $2M annual revenue or 100 active subscriptions (renamed from 'active contracts'), Growth still covers $2M-$20M with up to 1,000, Scale still applies above $20M. A company whose product IS usage-based billing crossed the usage/subscription line in both directions inside a week and settled back on the hybrid side. Nothing in this corpus argues harder that hybrid is an attractor rather than a waypoint.
- Wispr Flow · Aug 2026 — Counter-current at its strongest: an entire second PRODUCT absorbed into an unchanged subscription. Wispr shipped Wispr Notetaker — meeting transcription, speaker ID, cross-meeting Q&A, calendar and Slack connections, and MCP export into Claude and ChatGPT — as a new top-level nav item and help-center collection, included in Free and Pro at no extra seat price and marked 'Coming soon' on Enterprise. The site nav went from 'Dictation | Business | Pricing' to 'Dictation | Notetaker | Business | Pricing' and the footer now lists two products. No new SKU, no new meter, no price change. A vendor with an obvious per-minute metering opportunity (meeting transcription) chose to bundle it into a flat seat instead.
- Sequence · Jul 2026 — The anti-drift counter-current, stated at its sharpest: Sequence shipped named autonomous agents for AR, reconciliation, invoice review, quote approvals, payment reminders and contract intake, plus a Watchtower review console and Sequence MCP — and attached no meter to any of it. Growth stays $799/month, Core and Scale stay bespoke platform fees on projected billed revenue. A billing vendor added an agent layer that does metered work and absorbed it entirely into the existing subscription. Source: changes/sequence-2026-07-21-launch.md.
- Claude Code · Jul 2026 — Both directions in a single release, which is why the thesis has to be stated as an attractor rather than a trajectory: Anthropic absorbed three additional products (Cowork, Design, Science) into Claude Pro $20/mo, Max from $100/mo and Team $20-$100/seat/mo at UNCHANGED prices — more capability bundled into the subscription — while simultaneously adding a new metering axis, Managed Agents at $0.08 per session-hour, on top of per-token rates. Source: changes/claude-code-2026-07-23-packaging.md.
Trivia
-
Exa completed a full round-trip in the corpus's subscription-to-usage-drift arc: it dropped subscription tiers entirely for pure pay-as-you-go credits in July 2024, then re-introduced structured per-endpoint pricing cards on top of the PAYG base in April 2026 — going subscription → pure-PAYG → hybrid in sequence, ending at the same destination as vendors that never removed the subscription floor. Resemble AI and Jasper have since run the same arc, at wildly different speeds.
-
Hybrid pricing's emergence as the corpus's gravitational centre (65/158 at 41%, the largest paid-structure tag, behind only the freemium tag at 85) is a bottom-up convergence: no vendor described their model as "hybrid" when they built it. The label emerged from observing that the endpoint of both directions of drift — vendors leaving flat subscriptions and vendors walking back pure usage — is a seat-or-platform-fee plus a meter.
-
The bidirectional nature of the drift (subscriptions adding usage, pure-usage adding floors) means there is no default "more advanced" direction in AI pricing evolution. Bland AI re-coupling per-minute usage to subscription tiers (December 2025) and Exa re-adding endpoint cards (April 2026) both moved toward hybrid from opposite starting points — confirming hybrid as an attractor state rather than a midpoint on a one-way journey.
-
Hyperline crossed the usage/subscription line in both directions in SIX DAYS, and it sells usage-based billing for a living. On 2026-07-29 it dropped the percentage-of-billed-revenue take rate it had run since 2023 for a flat $599/month Launch tier — logged at the time as this trend's cleanest usage-to-subscription case. On 2026-08-04 Launch was back at $299/month plus 0.7% of revenue processed, with every eligibility band unchanged. It beats Resemble AI's six-week reversal by a factor of seven.
-
Chargebee is the first corpus vendor to sell the hybrid attractor as a toggle rather than embody it as a structure. Its single Flow plan (2026-08-11) is priced either at 0.80% of monthly billing value with a $0 platform fee, or at a $99/mo platform fee plus 0.65% — same product, same 100M usage-events allowance, buyer's choice of which pole to sit on. It also deleted the USD 250,000 free-billing threshold it had carried since at least 2022, so the percentage now applies from the customer's first dollar.
-
One structure tag lost companies while the corpus grew. Between the 353- and 380-company reviews, `commitment` fell from 63 companies to 58 — from 18% to 15% — the only pricing-structure tag to decline in absolute terms. Every other tag gained: freemium 169 to 177, subscription 160 to 173, hybrid 155 to 168, pure-usage 114 to 116, seat-based 73 to 78.
-
Resemble AI held the corpus record for the fastest packaging reversal until August 2026: it collapsed a five-tier subscription ladder into a single pay-as-you-go Flex plan in June 2026, then reintroduced Team ($280/mo annual, 5 seats) and Business ($800/mo annual, 20 seats) tiers above Flex on 2026-07-29 — six weeks later. It also cut the Flex detection rate from $0.04 to $0.035/sec and moved the sales nudge from $2,000/month of Flex spend down to $1,000/month on Business.
-
Jasper ran the same round-trip over three years instead of six weeks: its 2026-07-22 change entry opens "Jasper has reintroduced usage metering three years after killing the word caps it launched with." The new meter is granular and seat-adjacent rather than replacing the seat — 10 credits per Grid row or GEO Hub query, 40 per Research or Translation Agent run, 100 per Optimization Agent run, 1-40 per API call — on top of an unchanged $59/seat Pro price, with no dollar-per-credit rate published at all.
-
The pure archetypes have almost disappeared. Of 353 companies only 10 are tagged hybrid and nothing else, 20 subscription-only and 13 pure-usage-only, while 57 carry BOTH hybrid and subscription and 25 carry both pure-usage and subscription. Freemium (169), subscription (160) and hybrid (155) are now statistically tied for the lead — so the evidence for a hybrid attractor is the tag OVERLAP, not any single tag's rank.
For buyers
Pure-usage pricing is volatile to forecast, and many vendors walk it back toward hybrid for revenue predictability. Expect a base fee to reappear; negotiate the metered component and any minimums separately from the platform fee.
For vendors
Moving to usage needs metering and prepaid/credit plumbing; keeping revenue predictable then needs a platform fee or commitment on top. The stable destination is hybrid — a seat or base fee plus metered overage — so build for both from the start.
Outlook — what to watch
Hybrid will remain the centre of gravity. The interesting movement is at the edges: outcome/agent pricing (charge per resolved ticket or completed task) is the next experiment, but adoption is tiny so far. Watch whether agentic products can make a usage-only or outcome model stick where SaaS couldn't.
Bottom line
Several vendors went pure-PAYG (Exa, Tavus, ElevenLabs) then re-introduced structure. Hybrid (144/338, 43%) is the gravitational centre — the largest paid structure — not pure usage.
FAQ
Is usage-based pricing replacing subscriptions for AI?
Not wholesale. Several vendors moved from flat subscriptions to pure pay-as-you-go, but many then re-added tiers. The corpus shows hybrid — a base fee plus metered overage — as the stable endpoint, not pure usage.
Why do vendors move back from pure usage to hybrid?
Pure usage makes revenue hard to forecast for the vendor and spend hard to forecast for the buyer. A platform fee or commitment restores predictability while metering still captures heavy users.
What's the difference between usage-based and hybrid pricing?
Usage-based bills only for what you consume; hybrid combines a fixed recurring fee (often a seat) with variable usage charges. Hybrid is the most common paid structure in the corpus — 144 of 338 companies (43%) — behind only the freemium overlay (169).