The middle of the ladder churns — but the free tier is not the stable anchor this trend assumed
When an AI vendor reprices, the part that moves is almost always the paid self-serve middle — the $10–$200/mo Starter/Pro/Plus tiers. The free tier (an acquisition gate) and the enterprise tier (a 'call us') stay pinned, while the middle gets re-priced, re-metered, split, or deleted.
What's happening — and why
What's happening: pricing changes concentrate in the middle of the ladder. In the June 2026 batch, every self-serve repricing reworked a Starter, Pro, or Plus tier — WellSaid split its $50 plan into $10 + $33, Dust re-metered its Pro into Pro/Max, Runway re-metered its $95 tier, Tavus inserted a $22 Starter and a $975 Business — while free entries and custom enterprise tiers sat still (Dust even added a free tier in the same move).
Why: the two ends of the ladder are structurally pinned. The free tier is an acquisition decision and rarely moves; the enterprise tier is a private negotiation and never publishes a number. The middle is the only place a vendor is still solving for margin against moving model costs and unproven willingness-to-pay — so it absorbs all the experimentation.
How it works
Evidence over time
18 supporting · 7 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| WellSaid | Jun 2026 | Retired the single $50 Creative mid tier and split it into Starter $10 and Pro $33 — the paid middle of the ladder completely rebuilt while the free trial and Enterprise tier were untouched. |
| Dust | Jun 2026 | Re-cut the paid middle from a flat €29 Pro into credit-metered Pro ($30) and Max ($150) — and left the new free tier and custom enterprise alone. |
| Runway | Jun 2026 | Replaced the $95 Unlimited mid-upper tier with a credit-metered Max at the same price; the entry and enterprise ends of the ladder were not the part that moved. |
| Tavus | Jun 2026 | Reshuffled the middle — added a $22 Starter, renamed the $59 plan to Builder, inserted a $975 Business — while the free developer entry and custom enterprise stayed in place. |
| Windsurf | Apr 2026 | Raised the mid Pro tier $15→$20 and swapped its meter (credits→quotas); the free tier and the newly added $200 Max bracketed an unstable middle. |
| Synthesia | May 2026 | Cut self-serve plans ~34% with the entry/mid tier landing at $14 — the self-serve middle re-priced downward while enterprise stayed custom. |
| Ideogram | May 2026 | Re-set the mid band with a Plus $15 / Pro $42 grid — the paid self-serve tiers are where the new structure lives, between a free tier and a per-image API. |
| Galileo | Oct 2025 | Inserted a brand-new $100 Pro tier between the free plan and custom Enterprise — the mid band created where there had been a gap. |
| Synthflow | Jun 2026 | Resolved its unstable self-serve middle by deleting it — removed the per-minute PAYG path and went Enterprise-only, a terminal form of mid-tier instability. |
| Shortwave | Jun 2026 | Raised all four paid mid-band seats 20–29% (Pro $18, Business $30, Premier $45, Max $120) and dropped the annual toggle — the entire self-serve middle repriced in one move while the free tier and Enterprise sat still. |
| Creatify | Jun 2026 | More than doubled the Pro mid tier ($49→$99) and bumped Starter ($33→$39) while leaving the developer-API family and per-action credit table unchanged — the consumer self-serve middle was the only part touched. |
| Qodo | Jun 2026 | Terminal mid-tier instability, second form: deleted per-seat Teams pricing AND the permanent free Developer tier, collapsing the middle into a single pooled-credit Pro Team plan — the mid band re-architected entirely while Enterprise stayed custom. |
| Lokalise | Jun 2026 | Cut the Growth mid tier ($499→$375), trimmed its allowances, and removed the $0 Free plan — the paid middle re-priced while Explorer ($144) and Enterprise (quote-only) held. |
| Mem0 | Jul 2026 | Mid-band deletion, the surviving half of the thesis: retired the $79 Growth tier outright, collapsing the ladder to Hobby / Starter / Pro / Enterprise. The middle tier removed rather than repriced. |
| Freepik | Jul 2026 | The mid band rebuilt wholesale in one move: Magnific's plan ladder deleted Essential, repriced Premium+ and Pro, and added a new Business seat tier — a full reconstruction of the paid middle, not an adjustment. |
| Vellum | Jul 2026 | Mid band fragmenting rather than settling — the WellSaid pattern again: one Pro configurator (effective $50/mo minimum) became three named presets at $30, $100 and $200, with the a-la-carte path demoted to 'Custom'. Three mid-band SKUs where there was one. |
| Tavus | Jul 2026 | The fastest mid-band churn in the corpus: Tavus moved its 50%-off-first-month promo from Starter up to the Builder tier on 2026-07-21, then dropped the Builder promo entirely on 2026-07-29 — eight days later. The same mid-tier repriced twice in one week. |
| Poe | Jul 2026 | Mid-band devaluation with no price change: the Premium plan's monthly points allowance cut 34% while the headline price held — the mid tier repriced through its allowance rather than its number, which is the form of mid-band churn a price-diff cannot see. |
Counterexamples
- E2B · May 2026 — Its $150/mo Pro mid tier held identical across the entire Wayback record (2024-12 → 2026-05) — a stable middle, showing the churn is not universal.
- Intercom · Jun 2026 — Promo'd the entry Essential tier but left the mid/upper Advanced ($85) and Expert ($132) tiers and the Fin meter unchanged — here the middle held and the entry moved.
- Gumloop · Jul 2026 — Free-tier counterexample, and the cleanest falsification of the 'pinned acquisition gate' mechanism: Gumloop removed a long-standing $0/mo, 5,000-credit, 1-seat permanent Free plan and replaced it with a 14-day free trial of Pro, offered on MONTHLY billing only (the annual toggle shows a paid 'Select Plan' button instead of a trial CTA). The Pro credit slider itself ($37/20k credits up to $1,840/1M credits) and Enterprise were unchanged — so this vendor left the paid middle alone and repriced the free tier instead. The exact inverse of the trend.
- Gladia · Jul 2026 — Free-tier counterexample of the most consequential kind — a recurring allowance converted to a one-time grant while per-hour rates held: 10 hours of free transcription EVERY MONTH became 'a one-time grant with no monthly reset' of 50 EUR in credits, alongside a move to a prepaid wallet with auto top-up. Async $0.61/hr and real-time $0.75/hr both unchanged. A vendor can materially reprice its funnel without touching a single number on the paid ladder.
- Freshworks · Jul 2026 — Free-tier deletion at incumbent scale, with the paid mid-band left intact: the $0 three-user Freshsales plan was removed (only a 21-day no-credit-card trial remains) while Growth $9, Pro $39 and Enterprise $59 per user/month all held. In the same move it PUBLISHED previously unpriced AI add-ons — Freddy AI Agent at $49/100 bot sessions and Branded documents (CPQ) at $19/user/month. The monetization action was entirely at the free floor and in the add-on layer, not the middle.
- Socket · Jul 2026 — Free-tier counterexample in the OPPOSITE direction, same date as three others: Socket expanded its Free plan's tracked-dependency cap from 1,000 to 10,000 — a 10x increase — while Team ($25/dev/mo) and Business ($50/dev/mo) list prices held steady. The free tier moved and the mid band did not. Paired with Exa doubling its signup credit ($10→$20) and dropping the card requirement, and You.com adding a 100-calls/day free tier, all on 2026-07-29, this makes the free tier the most volatile band of that particular day.
- Inflection AI · Jul 2026 — Enterprise-end counterexample: Inflection retired its Enterprise offering entirely — /enterprise now returns a 404 — on the same day it launched Inflection AI Labs with Pi Journeys and two unnamed experiments. The 'stable contact-us tier' can be deleted as readily as a mid tier. Six other corpus vendors edited their enterprise end in the same 15-day window (runpod, abridge, fathom, langsmith, lago, 01-ai).
Trivia
-
Across two consecutive capture batches (2026-06-24 and 2026-06-30), every single self-serve repricing in the corpus landed in the $10–$200/mo middle band — 11 distinct vendors, zero free-tier-terms changes and zero enterprise-floor moves among them. The middle is not just the most-changed part of the ladder; in these two batches it was the *only* part that moved. That turned out to be a property of those batches, not of the corpus — the 2026-07-15 to 2026-07-30 window logged 12 free-tier edits and 7 enterprise-end edits alongside ~15 mid-band ones.
-
Every self-serve repricing in the 2026-06-24 capture batch touched the *middle* of the ladder — Starter / Pro / Plus tiers — while free and enterprise tiers were left intact. Dust even added a new free tier in the same change that re-cut its paid seats.
-
WellSaid deleted its only mid tier ($50 Creative) and replaced it with two ($10 Starter + $33 Pro) in a single 2026-06-24 change — the mid band fragmenting rather than holding.
-
Synthflow took the opposite exit on the same day: rather than re-cut its unstable self-serve middle, it deleted the self-serve path entirely and went Enterprise-only ($30K/yr floor). When the mid tier won't settle, one resolution is to remove it.
-
This trend wrote its own kill condition and then watched it fire. It said instability would be disproved as mid-specific if "free-tier terms or enterprise floors start moving as often as mid tiers." In the 15 days from 2026-07-15 to 2026-07-30 the score was roughly 15 mid-band edits, 12 free-tier edits and 7 enterprise-end edits. The middle is still the most active band — it is no longer the only one.
-
The free-tier moves in that window ran in BOTH directions on the same dates, which is the strongest evidence that the free tier is a priced lever and not an acquisition constant: on 2026-07-29 alone, Playground cut Free downloads from 10 to 2 per day while Socket 10x'd its free dependency cap from 1,000 to 10,000, Exa doubled its signup credit from $10 to $20 and dropped the card requirement, and You.com added a brand-new 100-calls-per-day free tier.
-
The most consequential free-tier edits were the ones that changed a RECURRING allowance into a one-time grant, because they look like generosity on the page: Gladia replaced 10 free transcription hours every month with a single non-resetting 50 EUR credit (2026-07-22), Cerebras replaced an open rate-limited Free tier with a one-time $5 credit (2026-07-21), and Puzzle turned "free until $20k of transaction volume" into "free for 2 months" while relabelling entry-tier AI credits as lifetime rather than recurring (2026-07-21).
For buyers
If you're a self-serve customer, you're sitting in the most volatile part of the ladder — the tier most likely to be restructured under you. Budget for it: assume your Pro/Plus plan's price, allowance, or meter can change at renewal, and prefer vendors that grandfather. A vendor still fragmenting or deleting its middle band hasn't found its self-serve unit economics yet.
For vendors
Mid-tier churn is a maturity tell — it signals you haven't settled willingness-to-pay in the self-serve segment. The two terminal moves are both visible in the corpus: fragment the middle into more tiers (WellSaid 1→2) so buyers self-select, or delete it and go enterprise-only (Synthflow) when the segment won't clear margin. Pick deliberately, and stabilise the free gate and enterprise floor so only the band you're actually tuning moves.
Outlook — what to watch
Logged emerging in June 2026. It confirms if repricing keeps landing on the $10–$200 band while free and enterprise tiers stay materially stable across the next cycles, and if more vendors resolve mid-tier instability by fragmenting or deleting it. It weakens if free-tier terms or enterprise floors start moving as often as mid tiers, or the middle stabilises as model costs flatten. Counterexamples already exist — E2B's $150 Pro held for 18 months — so it's a tendency, not a law.
Bottom line
The unstable part of an AI pricing ladder is the paid self-serve middle ($10–$200/mo). It gets re-priced, re-metered, split, or deleted while the free gate and the enterprise 'call us' tier stay pinned — because the middle is where willingness-to-pay is still being discovered.
FAQ
Which pricing tier changes most often at AI companies?
The paid self-serve middle — roughly $10–$200/mo Starter, Pro, and Plus tiers. In the June 2026 corpus batch every self-serve repricing reworked a mid tier, while free tiers and custom enterprise tiers were left intact.
Why don't free and enterprise tiers change as much?
They're structurally pinned. The free tier is an acquisition gate the vendor rarely touches, and the enterprise tier is a private negotiation with no published number to change. The middle is the only band where the vendor is still solving for margin and willingness-to-pay, so it absorbs the experimentation.
What does it mean if a vendor keeps changing its mid tier?
It usually means they haven't settled their self-serve unit economics. The two endgames are fragmenting the middle into more tiers so buyers self-select, or deleting it and going enterprise-only — both visible in the corpus the same week (WellSaid split its mid tier; Synthflow deleted its self-serve path entirely).