Expiring generosity: unlimited plans and promo rates with printed end dates
Three of 338 corpus vendors are time-boxing generosity: Genspark's Plus and Pro plans advertise unlimited SOTA-model usage 'only guaranteed through December 31, 2026,' Deepgram prints limited-time promotional streaming rates beside their struck-through standard prices, and Oxylabs runs a six-month 40% coupon (WU40) on Web Unblocker. This is an emerging signal, not an established pattern — the printed dates land from late 2026, and whether they convert into price increases, quietly renew, or get dropped is still open.
What's happening — and why
What's happening: a small set of vendors now publishes prices that announce their own end. Genspark's unlimited, zero-credit usage of 'the latest SOTA lineup' of chat and image models — the core promise of its Plus and Pro plans — carries an explicit sunset clause: only guaranteed through December 31, 2026 (verified 2026-06-02). Deepgram's pricing page (captured 2026-05-29) advertises 'limited-time promotional rates on streaming,' with the discounted streaming speech-to-text prices struck through against the originals, so the promo's reversal is pre-printed on the rate card. And Oxylabs (2026-06-04) put a six-month fuse on a 40% discount: checkout code WU40 cuts every Web Unblocker tier — Regular and Enterprise — by 40%, after which buyers revert to list. Unlike ordinary launch discounts, these clauses sit on the primary rate card and scope the core value promise: the unlimited allowance, the streaming rate, the per-tier price.
Why: the printed date hedges unpredictable model and infrastructure costs without committing to a list-price change. If model costs keep falling, the clauses quietly renew and become marketing texture; if costs rise — or usage of 'unlimited' tiers outruns margin — the dates are pre-authorized price increases that require no repricing announcement. The optionality is built into the page itself. It is far from a market default, though: E2B's headline prices are identical across its entire Wayback record (December 2024 through May 2026 — $150/mo Pro and unchanged per-second vCPU rates), and Dust's Pro plan has held at 29€/user/month since at least May 2024, through two funding rounds and a model-generation transition. Stability remains the dominant — and for some vendors deliberate — stance.
How it works
Evidence over time
16 supporting · 5 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Genspark | Jun 2026 | Plus and Pro advertise unlimited, zero-credit usage of 'the latest SOTA lineup' of chat and image models — but only guaranteed through December 31, 2026; the unlimited promise has an explicit sunset clause. Still unresolved as of this review — the corpus's longest-dated open guarantee. |
| Deepgram | May 2026 | Pricing page advertises 'limited-time promotional rates on streaming,' showing discounted streaming STT prices struck through against original rates — the promo's reversal is pre-printed on the rate card. |
| Oxylabs | Jun 2026 | Running 40% off Web Unblocker for six months via checkout code WU40, across all Regular and Enterprise tiers — a discount with a built-in expiration rather than a list-price cut. The six-month fuse lands early December 2026, still open at review. |
| Intercom | Jun 2026 | RESOLVED — reverted on schedule. The 'Save 35% on Essential plan' new-customer offer visible on the 2026-06-24 capture (Essential struck from $29 to $19/seat/mo, $348 → $228/yr) is absent from the 2026-06-30 capture; both the pricing page and the calculator show Essential back at flat $29/seat. Advanced ($85), Expert ($132) and the $0.99-per-Fin-outcome meter were never in scope and are unchanged. |
| Anthropic | Jul 2026 | Claude Sonnet 5 launched as the production-standard tier at an introductory $2/$10 per 1M tokens 'through August 31, 2026', reverting to standard $3/$15 on September 1 — a printed 50% increase on the flagship rate card, with batch following $1/$5 → $1.50/$7.50. Still displayed unchanged at the 2026-07-29 capture alongside Opus 5. The first time the mechanic attaches to core per-token pricing rather than to a discount badge. |
| Weights & Biases | Jul 2026 | Added ARIA as a distinct token-priced line on the pricing compare table, offered free 'for a limited time' on Free and Pro and included on Enterprise — a new metered product shipped with its own undated fuse. Core tiers (Free / Pro / Enterprise) and the storage, Weave-ingestion and inference meters were unchanged. |
| Tavus | Jul 2026 | Moved the 'first month 50% off' introductory discount one rung up the CVI developer ladder: Builder ($59/mo, 'Most Popular') began showing $59 struck through against a $29 first month in both the plan card and the Compare Plans header, while the $22 Starter tier lost the '50% OFF FIRST MONTH' badge it previously carried. |
| Puzzle | Jul 2026 | Converted a usage-gated free tier into a dated promo: the footnote 'Starter plan is free until $20k of transaction volume' was replaced by 'Starter is free for 2 months', with a struck-through $25 list price and a '100% off for 2 months' badge on the card. Starter and Core credits also relabelled '25 credits life time' (no monthly refill). No headline price moved — Starter $0, Core $30, Complete $50, Scale from $150. |
| Zhipu AI | Jul 2026 | RESOLVED — converted into a list increase. The GLM Coding Plan launched Sept 2025 at $3 for the first month, settled near $10/mo for Lite after that promo was withdrawn in February 2026, and now lists at $18 / $72 / $160 per month (Lite / Pro / Max) — roughly +26% effective at Lite and +40% at Max. The discount moved from a first-purchase promo into a published billing-term ladder (monthly -10%, quarterly -20%, yearly -30%), with the renewal rate printed on each card. |
| Braintrust | Jul 2026 | RESOLVED — quiet renewal. The 2026-07-22 capture said GLM-5.2 built-in-model access ran 'through July 31, 2026'; the 2026-07-30 capture removes that clause from both the billing FAQ and the Plans-and-limits docs, and folds GLM-5.2 into a renamed 'Model credits' pool with Topics. No rate changed. A second dated clause remains open on the same page: Pro's $249 included credit is printed as dropping to $100/month on September 1, 2026. |
| HubSpot | Jul 2026 | Launched Revenue Hub with HubSpot Payments at 2.9% + 0.5% per card transaction, carrying a 60-day no-platform-fee introductory promo; sellers who keep Stripe instead pay a flat 0.75% platform fee. A time-boxed fee waiver used to steer buyers onto the first-party payment rail. |
| DeepSeek | Aug 2026 | The dated-list-price form at frontier scale, and the arc ran in the open across three days. On 2026-08-11 the Models & Pricing page added a footnote distinct from its standing 'prices may vary' disclaimer: DeepSeek 'plan[s] to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly.' No amount, model or date. On 2026-08-14 it published the card — peak/off-peak billing effective 16:00 UTC on 2026-08-16, peak hours 01:00-04:00 and 06:00-10:00 UTC, all other hours off-peak at exactly half the peak rate. DeepSeek-V4-Flash cache-miss input moves from a flat $0.14 per 1M to $0.22 off-peak / $0.44 peak and output from $0.28 to $0.66 / $1.32; V4-Pro cache-miss input from $0.435 to $0.66 / $1.32. Roughly 1.5-2.5x off-peak and 3-12x at peak, printed in advance on a flagship per-token rate card, with no introductory rate on the way in. |
| Fireworks AI | Aug 2026 | A forward-dated increase across an entire hardware rate card, published as a two-column current-versus-future table. Every on-demand GPU tier rises effective 2026-09-01: H100/H200 $7.00 to $8.00/hr (+14%), B200 $10.00 to $13.00 (+30%), B300 $12.00 to $15.00 (+25%), GB300 $18.00 to $20.00 (+11%). Fireworks' own framing notes this is the first repricing of an already-published on-demand SKU since the product launched in January 2024 — every prior on-demand pricing event had been an addition rather than a change. The forward date functions as a notice period rather than a hedge: buyers get 20 days to commit at the old rate. |
| HubSpot | Aug 2026 | The fourth resolution type in the corpus — the promo LENGTHENS. HubSpot launched Revenue Hub on 2026-07-30 with 'no platform fee for 60 days' on HubSpot Payments; five days later the same copy on the same page reads 'no platform fee for 1 year' — a 6x extension confirmed on hubspot.com/pricing/revenue on both dates. Nothing else moved: the 2.9%+0.5% card rate, the 0.8%+0.5% ACH rate capped at $10, the flat 0.75% for sellers keeping Stripe, and the per-seat tiers (Free / Professional $85-95 / Enterprise $140) are identical to the 2026-07-30 capture. A dated waiver can be re-dated upward as easily as it can lapse. |
| FLORA | Aug 2026 | The fifth resolution type — expire, vanish, then revive with a new date. FLORA's Special launch offer added extra included generative usage worth $12/seat (Starter), $50/seat (Pro) and $100/seat (Max) per month; its stated deadline had already moved once from Jul 1 to Aug 1. After Aug 1 the badge disappeared entirely, confirmed missing across the 2026-08-11, 08-12 and 08-13 recaptures, so new and existing self-serve customers briefly got less included generation for the same seat price. On 2026-08-14 the identical offer was live again with 'Applies through Aug 31'. Base seat prices never moved ($18 Starter / $50 Pro / $200 Max), and FLORA still publishes no fixed base usage-pool figure per seat — so the only number a buyer can read is the one on a rolling monthly deadline. |
| Anthropic | Aug 2026 | RESOLVED — cancellation, the sixth resolution type and the first at frontier scale. Claude Sonnet 5 launched in July 2026 at an introductory $2/$10 per 1M input/output with standard pricing of $3/$15 explicitly scheduled for September 1, 2026 — the printed 50% increase this trend named as its live test, and a return to the $3/$15 Sonnet anchor held since Claude 3. On 2026-08-28, three days before the date, Anthropic cancelled it: the API pricing docs now state $2/$10 is the permanent standard price, the September 1 increase will not occur, and Batch API pricing is correspondingly fixed at $1/$5 with no scheduled rise to $1.50/$7.50. A dated clause on the corpus's most-cited rate card was withdrawn in the buyer's favour, which no prior resolution had done. |
Counterexamples
- E2B · May 2026 — Headline prices identical across the entire Wayback record (2024-12 through 2026-05): $150/mo Pro and unchanged per-second vCPU rates — durable list pricing with no promo mechanics at all.
- Dust · Jun 2026 — Pro has held at 29€/user/month since at least May 2024, through two funding rounds and a model-generation transition — the opposite stance: stability as the signal.
- Kaiber · Jul 2026 — Ran the mechanic in reverse: the Help Center now states credits 'don't expire and roll over month to month', replacing an earlier use-it-or-lose-it cycle reset. Plan prices held ($10 / $29 / $99 monthly). De-timing generosity, not time-boxing it.
- Hyperbolic · Jul 2026 — Formalised prepaid compute credits that are '1:1 with dollars' and 'never expire', with a $5 minimum and Auto Top-Up, alongside a self-serve 1-week-to-1-month reserved-capacity ladder. Explicit permanence written into the money mechanics.
- Playground · Jul 2026 — Moved the other way on promo infrastructure: removed the coupon-code / 'Apply promo code' field entirely from its Pro and Pro Plus checkout cards while holding list prices ($15/mo Pro, $45/mo Pro Plus). Cutting the discount plumbing out rather than adding a dated offer.
Trivia
-
The corpus's most-watched printed price increase never happened. Anthropic had published Claude Sonnet 5 at an introductory $2/$10 per 1M "through August 31, 2026," with a standard $3/$15 scheduled for September 1 — a printed 50% rise on a flagship model. On 2026-08-28, three days before the date, Anthropic cancelled the increase and made $2/$10 permanent, breaking a $3/$15 Sonnet anchor that had held since Claude 3.
-
Ten dated expiries have now resolved in the corpus and they have produced six different endings — quiet renewal, on-schedule reversion, conversion into a permanently higher list, extension, expire-then-revive, and cancellation. Not one ending has repeated at frontier scale, which is why the printed date should be read as an option the vendor holds rather than a scheduled price.
-
Genspark's Plus and Pro plans (verified 2026-06-02) advertise truly unlimited, zero-credit usage of top-tier chat and image models — "only guaranteed through December 31, 2026." The word "unlimited" ships with a printed expiry roughly seven months out, an explicit hedge against model-cost drift.
-
Deepgram's pricing page (2026-05-29 capture) shows "limited-time promotional rates on streaming" with the discounted streaming STT prices struck through against the originals — a rate card that displays its own future price increase.
-
Oxylabs (2026-06-04) put a six-month fuse on a 40% discount: coupon WU40 cuts every Web Unblocker tier by 40% for exactly six months, after which buyers revert to list — a time-boxed land-grab in a category (unblocking/scraping) already in price deflation.
-
Braintrust deleted its own deadline with one day to spare. The 2026-07-22 capture said GLM-5.2 built-in access ran "through July 31, 2026"; the 2026-07-30 capture has the clause gone entirely from both the billing FAQ and the Plans-and-limits docs. It is the first printed expiry in the corpus to actually arrive — and it resolved by quiet renewal, not by the price increase the clause reserved the right to make.
-
Tavus ran the shortest-lived promo in the corpus: it moved its "first month 50% off" badge from the $22 Starter tier up to the $59 "Most Popular" Builder tier on 2026-07-21, then deleted it entirely on 2026-07-29 — eight days. No list price, included-minute allowance or overage rate changed in either edit; only the badge moved.
-
Anthropic's Sonnet 5 launch (2026-07-06) is a printed 50% price increase: $2/$10 per 1M tokens through 2026-08-31, then $3/$15 from September 1. That makes the largest lab in the corpus the first vendor to put the expiring-generosity mechanic on flagship per-token rates rather than on a discount badge — and buyers on the intro rate have a dated, pre-authorized COGS step-up they cannot negotiate away.
-
Puzzle (2026-07-21) shows the mechanic eating a real entitlement, not just a discount: its footnote "Starter plan is free until $20k of transaction volume" — a usage gate a pre-revenue company could sit behind indefinitely — was replaced by "Starter is free for 2 months" plus a struck-through $25 and a "100% off for 2 months" badge. A volume-based free tier became a two-month trial with no price change on any card.
For buyers
Read the date as part of the price. If a plan's core promise carries a sunset — Genspark's unlimited usage is only guaranteed through December 31, 2026 — budget and contract as if the post-expiry terms were unknown, because they are: the vendor has reserved the right to reprice without an announcement. Where the reversal is printed, model your steady-state cost at the undiscounted number: Deepgram's struck-through standard streaming rates tell you exactly what the promo reverts to, and Oxylabs' WU40 coupon lapses to list after six months, so a 40% discount should be amortized, not extrapolated. And note that durable pricing is still on the market: E2B's headline prices are unchanged across its entire Wayback record and Dust has sold the same 29€/user/month since May 2024. If predictability matters more than the promo, the stable vendors are the comparison set.
For vendors
Time-boxing generosity takes more than a marketing banner. The date has to be printed where the buyer reads the price — Genspark puts the guarantee window in the plan copy itself — and ideally the post-promo price stays visible, the way Deepgram strikes through its standard streaming rates next to the discounted ones, which turns a future increase from a surprise into a disclosed term. Coupon mechanics need a built-in lapse (Oxylabs' WU40 reverts every Web Unblocker tier to list after exactly six months), which means billing that supports scheduled rate reversion and a grandfathering decision made before the date, not at it. The trade-off is trust: a printed expiry is honest about cost uncertainty, but it also tells buyers the price is provisional — the opposite signal from Dust, which has held 29€/user/month since at least May 2024 and treats stability itself as the pitch.
Outlook — what to watch
Logged as emerging in June 2026 at a corpus of 207 (now 338), on three evidence points — too few, and too recent, to call a pattern. The real test is already scheduled: the printed dates land from late 2026, starting with Genspark's December 31, 2026 sunset on unlimited usage. The signal is confirmed if a fourth vendor adds dated guarantees, or if Genspark's sunset converts into a metered or higher-priced plan; it dies if the dates pass with guarantees renewed indefinitely and no new adopters — in which case this was promo copy, not a pricing mechanism. With E2B and Dust demonstrating the opposite stance at no apparent cost, the burden of proof sits on the expiry clause.
Bottom line
Three of 338 corpus vendors now print expiry dates on their own generosity — Genspark's unlimited guarantee ends December 31, 2026, Deepgram's streaming promo displays its own reversal, and Oxylabs' 40% coupon lapses after six months. It's an emerging signal with a built-in verdict date: watch what happens when the clauses come due in late 2026.
FAQ
What is an expiring price guarantee?
A published price, discount, or plan promise that carries an explicit end date on the rate card itself. Corpus examples: Genspark's unlimited, zero-credit usage of top-tier chat and image models 'only guaranteed through December 31, 2026'; Deepgram's limited-time promotional streaming rates shown struck through against the standard prices; and Oxylabs' WU40 coupon, which cuts every Web Unblocker tier 40% for exactly six months before reverting to list.
Is Genspark's unlimited plan really unlimited?
The usage is advertised as unlimited and zero-credit across 'the latest SOTA lineup' of chat and image models on the Plus and Pro plans — but the promise itself is time-boxed. As of the 2026-06-02 verification, it is only guaranteed through December 31, 2026; what happens after that date — renewal, metering, or a higher price — has not been announced.
Why would a vendor print an expiry date on its own pricing?
It hedges unpredictable model and infrastructure costs without committing to a list-price change. If costs fall, the clause quietly renews; if costs rise or usage of 'unlimited' tiers outruns margin, the printed date works as a pre-authorized price increase that requires no repricing announcement. Either way, the optionality is built into the page itself.
Do most AI vendors time-box their pricing this way?
No — 3 of the 338 companies in the corpus do, which is why the trend is classified as emerging rather than established. The opposite stance is common: E2B's headline prices are identical across its entire Wayback record (December 2024 through May 2026, including the $150/mo Pro plan and unchanged per-second vCPU rates), and Dust's Pro plan has held at 29€/user/month since at least May 2024, through two funding rounds and a model-generation transition.