Self-serve list-price inflation

What's happening — and why

How it works

Evidence over time

36 supporting · 14 counter — hover or tap a point for detail, click to jump to the row.

Evidence

Trivia

• Lokalise raised its Growth plan by exactly $4 a month (from $375 to $379) on 2026-09-20. Three months earlier, on 2026-06-30, it had cut that same plan to $375.

• Beautiful.ai's Teams plan still costs $40 a month after 2026-09-11, but that price now covers 3 users rather than 1, while its single-user Pro plan rose 21%.

• Fireworks raised the price of something it had already published for the first time in two and a half years. Its on-demand GPU product launched in January 2024, and every pricing event since had been the ADDITION of a new tier — until 2026-08-12, when it posted a forward-dated increase across the entire card effective 2026-09-01 (H100/H200 $7→$8, B200 $10→$13, B300 $12→$15, GB300 $18→$20). It landed three weeks after a $1.5B Series D at a $17.5B valuation and an announced $1B ARR.

• Hyperbolic's H100 SXM went from $1.50/hr in June 2026 to $3.19/hr in August — up 113% in about two months across three consecutive captures — and in the same edit the company deleted its "starting at $0.20/GPU/hr" banner, which had never had a bookable card behind it, replacing it with "$3.19", the H100 rate itself. The advertised floor and the cheapest real price finally agree.

• A published price cut can expire. SambaNova restored gemma-4-31B-it to $0.38/$1.15 per 1M on 2026-08-11, its pre-cut June rate, less than three weeks after cutting it to $0.22/$0.59 — a 73% input and 95% output increase on an unchanged SKU. Hyperline's flat $599/month Launch tier lasted about a week before reverting to $299 + 0.7% of revenue. Neither shows up as a raise in any year-over-year comparison, because the discount is gone from the record too.

• Make held its price at $9/mo and halved what it buys. On 2026-08-11 it collapsed the Core/Pro/Teams ladder it had built on 2026-06-11 back into a single "Make Plan" at the same $9/mo annual headline, with the credit slider's entry step cut from 10,000 to 5,000 credits per month. The cleanup was still live on the page: an FAQ referencing "the Core, Pro, and Teams plans", an Enterprise card reading "Everything in Teams, plus:", and a pricing card and comparison table that disagreed on whether the surviving plan is called "Make Plan" or "Make Pro".

• Novita emptied its free-model shelf twice in a month, and the second free listing lasted three days. Three "Time Limited Free" LLMs converted to paid on 2026-08-11 (Macaron V1 Venti to $1.5/$4.5 per 1M, Macaron V1 Tall to $0.45/$2.6, Ling 3.0 Flash to $0.06/$0.18); Ling 3.0 Tiny was added at $0 the same day as the replacement, and was delisted outright on 2026-08-14 — the shortest-lived free listing ever tracked on that rate card.

• Mistral's 2026-07-06 API hike is the corpus's first frontier-lab self-serve *increase*: Small 4 rose 50% on input and 100% on output ($0.1/$0.3 → $0.15/$0.6 per 1M) and the OCR model doubled to $4/1K pages — all while the Vibe consumer ladder (Free/$14.99/$24.99) and the flagship Medium 3.5 / Large 3 rates held flat. The raise landed on the developer meter, not the human-facing subscription, inverting the usual scissors where consumer ceilings rise and token rates fall.

• Intercom raised its entry price on 2026-06-30 without changing a single list number: its 35%-off Essential new-customer promo simply expired, resetting the entry seat from $19 back to the $29 list ($228 → $348/yr). Promo withdrawal is the quietest form of self-serve inflation in the corpus — the sticker never moved, but new buyers paid 53% more than they would have a week earlier — until the discount came back on 2026-07-21.

• On a single day — 2026-06-30 — three corpus vendors raised self-serve list prices while three others (Together AI, Novita, Vercel) cut them: Creatify more than doubled its Pro plan ($49→$99) the same week Together AI cut its on-demand H100 floor twice in seven days. The app layer and the infra layer moved in opposite directions on the same date — the cleanest single-day illustration of the app-up/infra-down scissors. It did not hold: by late July Hyperbolic and Together AI were raising GPU rates, and by September on-demand GPU list prices had risen at eight vendors.

• lemlist (2026-06-30) raised its Email plan 77% ($39→$69) but lifted the included send allowance 10x (5,000→50,000 emails/mo) — the hike and the allowance bump together mean a high-volume sender's effective per-email price actually fell, while a low-volume sender's bill nearly doubled. The repricing is a deliberate re-segmentation disguised as a price increase.

• Creatify (2026-06-30) is the corpus's sharpest example of a hike with no value added: both Starter and Pro kept identical credit allowances (100 / 300 credits/mo) while Pro's price doubled — a straight margin grab, confirmed by the simultaneous cut of the annual discount from 'up to 50%' to 'up to 15%'.

• Eight of the biggest effective increases in the 2026-07-15 → 2026-07-30 window moved no list price at all. Poe held $19.99 while cutting its point allowance 34%; Composio cut Pro's included tool calls from 200,000 to 50,000 at an unchanged $29; Fathom pulled SSO out of Team; LangSmith moved Email Support off the $39 Plus plan to Enterprise-only; Arize deleted its $10/M-spans and $3/GB overage rows; Playground cut free downloads 10 → 2/day; Freepik cut its annual discount from ~40% to 25%; and Cartesia deleted its Monthly/Yearly toggle so advertised rates rose from $4/$39/$239 to $5/$49/$299. Reading list prices alone misses most of 2026's self-serve inflation.

• Cartesia raised prices ~25% by deleting a UI control. Through mid-2026 its plan cards sat inside a panel with a Monthly / Yearly switch where Yearly was badged "Save 20%" and rendered as the DEFAULT view — so the advertised prices were $4, $39 and $239. The July page has no toggle and shows $5, $49 and $299. The comparison grid still carries the vestigial line "same billing period as above", the only remaining trace of the control, and every credit allowance (20K / 100K / 1.25M / 8M per month) is unchanged.

• The app-up/infra-down scissors inverted inside one window. Hyperbolic nearly DOUBLED its published H100 SXM starting rate ($1.50 → $2.89/hr, +93%, with H200 +45% and B200 +71%) and pulled its consumer-GPU rates entirely, and Together AI raised reserved H100 tenors for the first time since its June cuts — while at the app layer Juicebox CUT Starter 14% ($139 → $119) and doubled its bundled credits, and Intercom revived the Essential discount on both cadences. Both layers now move both ways.

• Intercom reversed its own evidence in three weeks. The 2026-06-30 review logged its expiring 35%-off Essential promo as "the quietest form of self-serve inflation" — a 53% effective rise with no list number changed. On 2026-07-21 the discount came back WIDER: $29 struck to $19/seat/mo on annual ("Save 35%") and, new this time, $39 struck to $29 monthly ("Save 25%"). It ships inside a container tagged `data-experiment="team-self-serve-essential-price-reduction"`, so the entry price is an allocated A/B test — the promo lever is a dial, not a ratchet.

• Tavus ran the promo lever in both directions in eight days: on 2026-07-21 it moved its "50% off first month" badge UP from the $22 Starter tier to the Most Popular $59 Builder tier (showing $59 struck against $29), then on 2026-07-29 removed the discount entirely, leaving a plain $59/mo and no promo badge anywhere on the page. No list price, included-minute allowance or overage rate changed in either move — the same buyer saw three different first-month prices in nine days.

For buyers

If you are on a self-serve AI plan, mid-2026 is a repricing season — audit your renewal before it lapses. Watch for hikes disguised as repackages: a larger credit pool or a higher allowance can still raise your effective rate if your usage sits below the new bundle (lemlist's 10x allowance bump nearly doubled a low-volume sender's bill while cutting a heavy sender's per-email cost). Lock annual pricing before the discount shrinks — Creatify cut its annual discount from 'up to 50%' to 'up to 15%' the same day it doubled Pro, and Shortwave dropped the annual toggle entirely. In low-switching-cost categories — email clients, creative tools, outbound — keep a cheaper rival benchmarked, because that is exactly where a coordinated hike invites churn.

For vendors

Running this play needs a metering layer flexible enough to pair a list-price hike with a credit-pool or allowance change in the same release, so the increase lands as a re-segmentation rather than a naked price grab. The reusable move is to raise the headline while raising the included allowance even faster (lemlist) or swapping unlimited fair-use for a metered envelope (Dust) — protecting heavy users while lifting the floor on light ones. The risk is concentration: hiking in a low-switching-cost category, or pairing the hike with a compressed annual discount that removes the lock-in incentive, exposes you to churn toward a rival that held its price. The counter-move some vendors chose — Tavus added included minutes at the same price, Hume held all tiers and shipped a cheaper model toggle — shows the inflation is not obligatory even within self-serve.

Outlook — what to watch

Logged as new in June 2026 on a single-batch frequency cluster — six self-serve app hikes against four counterexamples, not a corpus-wide law. It sharpens into a confirmed pattern if the next change batches show more app-layer vendors repricing upward while infra rates keep falling, widening the spread the thesis predicts. It weakens if the hiking cohort sees visible churn and rolls prices back, or if more self-serve vendors follow Tavus and Hume and add value at flat prices instead. The thing to watch is the divergence itself: as long as the app layer raises list prices in the same windows the infra layer cuts, the 'margin migrating up the stack' reading holds.

FAQ

Related trends

• Annual prepay discount compression — a repricing tactic, not a category direction
• Token prices fall as access tiers rise
• The middle of the ladder churns — but the free tier is not the stable anchor this trend assumed