One vendor, several prices: first-party surfaces disagree
Six corpus vendors published two or more conflicting live prices for the same product across their own surfaces inside 19 days — five of them in a single four-day stretch. Intercom split three ways on one seat price: its pricing page said $29 while the calculator embedded in that same page, and its fin.ai sub-brand, both said $19.
What's happening — and why
What's happening: a vendor's list price is increasingly not one number but a set of numbers that disagree with each other, at the same moment, on surfaces the vendor owns. A marketing pricing page, a developer-docs billing reference, an in-page calculator, a navigation caption, a sub-brand domain and a marketplace listing can all assert a different rate for the same product.
Why: release lag plus surface proliferation, not deception. Docs are versioned with the API and move when the billing code moves; the marketing page moves when marketing ships. Any vendor whose meter changes faster than its slowest surface runs a disagreement for the length of the gap — and meters are changing fast, with 87 material changes landing across 69 corpus companies in the 16 days to 2026-09-01.
The part that makes it a buyer problem is that the disagreement has no reliable direction. Marketing is not dependably the optimistic surface, and the freshly-synced surface is not dependably the durable one.
How it works
Evidence over time
6 supporting · 3 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Intercom | Aug 2026 | One day after deliberately syncing three surfaces at $19/seat/mo, intercom.com/pricing's annual toggle reverted Essential to the clean $29/seat/mo list rate with no discount banner, while the site's OWN embedded calculator and the fin.ai/pricing sub-brand both continued to show $19/seat/mo (Save 35%, $228/yr). Three first-party surfaces, two prices, same day — and the first recorded split between two surfaces on the same domain. |
| Sarvam AI | Aug 2026 | A full marketing-pricing rebuild closed one gap and opened three. The month-old ~7x LLM price gap between marketing and the developer docs resolved in the DOCS' favour (Sarvam-105B at INR 29.28 / 10.98 / 73.2 per 1M, with the deprecated Sarvam-30B removed from marketing entirely). But marketing's new doc-translation line is tiered at INR 0.004-0.005 per character (about INR 40-50 per 10K) against the docs' unchanged flat INR 20 per 10K; marketing's new per-second dubbing table (INR 1.67 down to INR 1.20/sec) does not map cleanly onto the docs' per-minute editor_flow table; and a new Extraction API at INR 1.00/page appears on marketing only. |
| Linkup | Aug 2026 | Billing docs split the Fetch endpoint from two prices to five — mode (standard/pro) x renderJs, plus an optional flat $0.001 structured-output surcharge — pushing the top rate from $0.005 to $0.011. The marketing pricing page continued to advertise only the old $0.001-$0.005 range, so the docs surface is the expensive one. |
| Bright Data | Aug 2026 | Browser API moved off a fixed four-tier GB ladder onto Free / PAYG ($8/GB) / Scale ($499/mo, slider-adjustable) / Enterprise. The site navigation still carried a $5/GB "starts from" caption while the discrete $6/GB and $999 / $1,999 breakpoints were no longer statically shown anywhere on the product page — the nav asserts a floor the pricing page no longer documents. |
| Moonshot AI | Aug 2026 | A new Business tab on kimi.com/membership/pricing exposed a self-serve $600/year/seat price (minimum 2 seats) with a working seat-stepper and Subscribe checkout, three weeks after the tier was pulled sales-only — while the dedicated Kimi Business marketing page continued to show "Contact sales" only. Both motions coexist on first-party surfaces, so whether the product is self-serve depends on which page the buyer lands on. |
| Groq | Aug 2026 | groq.com/pricing 308-redirected to the homepage while every token, hour and character rate survived byte-identical in Groq's own developer docs — the page and the price parted company. Only the three built-in agentic tools genuinely lost a public rate, because their docs entries pointed at the deleted page rather than restating the numbers. |
Counterexamples
- AssemblyAI · Aug 2026 — Moved decisively the other way: itemized the $4.50/hr Voice Agent bundle into three component rows (Universal-3.5 Pro Realtime STT, Voice Agent LLM, Voice Agent TTS) all marked "Included", and added an FAQ explicitly stating there are no per-layer add-ons, concurrency fees or per-agent subscriptions, and no markup on BYO-Twilio SIP trunking. Two surfaces deliberately made to say the same thing, with the headline rate unchanged.
- Cerebras · Aug 2026 — Removed ZAI-GLM-4.7 ($2.25/$2.75 per 1M) from the public rate card AND the docs Model Catalog on the same day — the exact deprecation date it had footnoted in advance. Two surfaces moved in lockstep on a pre-announced schedule, which is what the absence of this trend looks like.
- Anthropic · Aug 2026 — Cancelled Claude Sonnet 5's scheduled 2026-09-01 increase and updated the API pricing docs to state that $2/$10 per 1M is the standard price, with Batch correspondingly fixed at $1/$5. The announcement and the docs agreed on the day of the decision, on the corpus's most-cited rate card.
Trivia
-
Intercom ran a three-way disagreement with itself on a single seat price. On 2026-08-26 it synced all three surfaces to $19/seat/mo — the first time this cycle that fin.ai/pricing matched the main site. One day later, on 2026-08-27, intercom.com/pricing reverted to the $29 list while the calculator embedded in that same page, and fin.ai, both kept showing $19. The disagreement was created by the fix.
-
Linkup inverts the usual assumption that marketing inflates and docs deflate. Its billing docs added a second pricing dimension (a "pro" Fetch mode) plus a flat $0.001 structured-output surcharge on 2026-08-28, taking the top Fetch rate from $0.005 to $0.011 — a 2.2x increase — while the marketing pricing page still advertised the old $0.001-$0.005 range. The buyer reading the friendlier surface is the one who under-budgets.
-
Groq is the cleanest proof that a dead pricing page no longer means a dark price. It 308-redirected groq.com/pricing to its homepage on 2026-08-11 while every token, hour and character rate survived byte-identical in its own developer docs. The only rates that actually went dark were the three built-in agentic tools — because their docs pointed at the page that had just been deleted.
For buyers
Stop citing a price without citing the surface it came from. Before you sign, pull the rate from the developer or billing docs as well as the marketing page — and if they disagree, get the one you intend to be billed at written into the contract. Do not assume the cheaper surface is the accurate one: at Linkup the docs carry the higher number (a top Fetch rate of $0.011 against a marketing page still advertising $0.001–$0.005), so the buyer reading the friendlier page is the one who under-budgets. If a vendor operates a sub-brand or a marketplace listing, check that too; for Observe.AI and Poolside the marketplace listing is the only place any price is published at all.
For vendors
This is a release-engineering problem wearing a pricing costume. If the meter is defined in code and the marketing page is defined in a CMS, they will drift, and the drift will be discovered by a prospect mid-evaluation. The fix is a single source of truth for rates that every surface renders from — pricing page, in-page calculator, docs, and any sub-brand — so a rate change is one edit rather than four. If you cannot do that yet, at minimum make the docs authoritative and say so on the marketing page. Note also that syncing is not a one-time task: Intercom got all three surfaces to agree and lost the agreement the next day.
Outlook — what to watch
Expect this to get worse before it gets better, because both drivers are still growing: vendors keep adding surfaces (calculators, sub-brands, marketplace listings, AI gateways that resell the same rate) and meters keep changing faster. The trend is falsified if disagreements resolve within one release cycle in most cases — making this latency rather than structure — or if new cases turn out to be deliberate segmented offers, a genuinely different price for a genuinely different buyer, rather than two surfaces both claiming to state the same list price. Watch whether any vendor ships a published single-source rate card; that would be the counter-move.
Bottom line
Six vendors ran conflicting live prices across their own surfaces in 19 days, and the direction of the error is unpredictable. The list price is no longer a citation — it is a query that needs a surface attached, and the answer can change depending on which page you land on.
FAQ
Which price is the real one when a vendor's pages disagree?
There is no general rule, which is the finding. At Linkup the developer docs carried the higher and newer rate ($0.011 top Fetch) while the marketing page still showed the old $0.001–$0.005 range; at Intercom the marketing pricing page carried the higher $29 list while its own embedded calculator and its fin.ai sub-brand both showed $19. Get the number you intend to be billed at into the contract rather than inferring which surface is authoritative.
Is this the same as a vendor hiding its pricing?
No. Hiding a price is covered by pricing-page-withdrawal, and changing a price over time is covered by pricing-page-as-experiment-surface. This trend is the third case those two miss: two live surfaces, the same moment, different numbers. Groq is the clean illustration of the difference — it redirected its pricing page to its homepage while every rate survived byte-identical in its developer docs, so the page went dark but the price did not.
Why would a vendor let its own pages disagree?
Because the surfaces have different owners and release cycles. Docs version with the API and move when billing code moves; marketing pages move when marketing ships. Sarvam AI's 2026-08-28 rebuild is the clearest example — it closed a month-old roughly 7x LLM price gap with its own docs while simultaneously opening three new gaps on translation, dubbing and a marketing-only Extraction API.
What does the absence of this look like?
Deliberate reconciliation. AssemblyAI itemised its $4.50/hr Voice Agent bundle across its pricing table and added an FAQ confirming there are no per-layer add-ons; Cerebras removed a deprecated model from its rate card and its docs catalog on the same pre-announced date; Anthropic cancelled a scheduled increase and updated its API docs the same day. In each case two surfaces were made to say one thing on purpose.