Blog · September 3, 2026
Pricing is usually the least engineered part of an engineering company. It gets decided in a spreadsheet, presented as tiers, and defended with adjectives. We would rather show the arithmetic, because the arithmetic is the whole idea.
Starting today, every paid call on Nyquest is priced by one rule. The engine compresses your context before the provider bills for it. Whatever that compression saves against the provider's own rate is the diff. We split it down the middle: your half is credited straight back to your wallet, on the same transaction as the charge; the other half is our fee. There is nothing else on the bill.
We tried the obvious things on paper first, and each one pays us for the wrong behavior.
A subscription pays us the same whether the engine saved you 75% or 0% this month. It also means the first thing a new user does is decide whether they trust us enough to pay before they have seen a single number. Most people, sensibly, do not.
A markup on tokens is worse, and it is what most routing layers do. A percentage on top of the provider's rate pays us more when we send more tokens. For a company whose entire product is sending fewer tokens, that is not a pricing model. It is a conflict of interest with a login page.
The split is the only structure we found where our revenue and your savings are the same number, measured on the same call. If the engine does nothing useful on a request, the diff is zero, and half of nothing is nothing.
Take a team's month on a frontier-class model at $3.00 per million input tokens: 100 million prompt tokens, mostly long conversations with plenty of history — the kind that compresses well. Say the engine removes 40% of those tokens before sending.
| Line | Tokens | At $3.00 / 1M |
|---|---|---|
| Provider rate for the month, uncompressed | 100,000,000 | $300.00 |
| Tokens compression removed — the diff | 40,000,000 | $120.00 |
| Your half of the diff, credited back to your wallet | +$60.00 | |
| Net cost to you | $240.00 | |
| What the provider actually charged for 60 million tokens | 60,000,000 | $180.00 |
| Nyquest's half — our fee, the only one | $60.00 |
Two things to notice. Net, the team paid $240 for $300 of provider-rate usage — 80% — on traffic where compression removed 40% of the tokens; the split is why 40 turns into 20. And the $60 that came back is exactly our $60 fee, which is the point: neither number exists without the other.
Now the other case. A month of nothing but short, history-free questions — 10 million tokens of one-liners. The engine looks at each one, finds nothing to remove, and sends it as-is.
| Line | Tokens | At $3.00 / 1M |
|---|---|---|
| Provider rate for the month | 10,000,000 | $30.00 |
| The diff | 0 | $0.00 |
| Credited back to you | $0.00 | |
| Net cost to you | $30.00 | |
| Nyquest's fee | $0.00 |
You paid the provider's rate. Not the provider's rate plus something — the rate. We made nothing on the month, and we think that is correct. A company that charges for compression should not be paid on traffic it could not compress.
The homepage says 50/50 and stops there, because 50/50 is a pricing term, not a measurement. The measurements live here. Two numbers we published in the compression week in review: a 33% average token saving on live production traffic over a 30-day window, and 75% on long conversations, on a curve that starts at about 1% at five messages and reaches 75% by nineteen.
Run the split over those and the bill becomes simple to predict. At the live-traffic average, you net about 83.5% of the provider's rate. On a long conversation at the top of the curve, about 62.5%. On the first message of a new chat, roughly 100%, because there is nothing to compress yet — the saving compounds as you work, and so does the credit.
In wallet terms: a $100 fund buys about $120 of provider-rate usage at the live-traffic average, and about $160 at the top of the curve. Credits land on every compressed request, so the balance drains slower than the provider's rate says it should, and the gap widens as the conversation grows. You get more, and you save more, from the same hundred dollars.
Per request, at the rate of the model that actually served it. The diff is the number of tokens the compressor removed, priced at that rate, and it is computed when the request is charged — not estimated at month end. Your half is credited to your wallet in the same transaction as the request's charge, so the two numbers land together and can be checked against each other.
The Savings page in the app keeps the ledger. It shows tokens saved for the period as a percentage of your prompt tokens, your share credited against what was charged, the split by model, and a list of recent requests as prompt → sent tokens with the credit for each. Today and all-time totals sit at the top. The arithmetic above is all there is to it; the page just does it for every request.
One floor, and the page marks it when it applies: a request never costs less than what the provider charged us for it. That usually happens when a request carries large memory or web-grounding context you were not billed for. On those, the credit is held back to the floor rather than paid out of our pocket — which is the same rule as the rest of the post, read from the other side.
Rates are the provider's rates. We do not mark up the base, and when a provider changes its price the change flows through unchanged. The model picker shows the live rate before you send, as it always has.
Free is $0: bring your own keys to any provider and the provider bills you directly, plus whichever models in the catalog happen to be free at the time. Nyquest charges nothing here, and BYOK sits outside the split entirely — your provider, your bill.
Pro is one $5 fund, once. It unlocks all 418 hosted models permanently, raises the limits, and after that every wallet-billed request credits your half of the diff straight back to the balance. There is no monthly fee and nothing renews; when the balance runs down you are still Pro, you just have an empty wallet.
Enterprise is the same split at your volume, with custom agreements, limits, and support.
That is the part we are proudest of. Whether you fund $5 or negotiate a contract, the rule is the same rule, and the split is the only thing Nyquest charges. Nobody has a worse deal for being small.
Being precise about the claim. The engine is unchanged by this post — this is a pricing change, not a compression release, and the savings curve is exactly what it was last week. BYOK is still free and still billed by your provider. Free models are still free. Rate limits, file limits, and the $5 Pro activation are what they were. The only thing that moved is the shape of the bill: from a rate with the margin somewhere you couldn't see, to a charge and a credit in the same transaction, with the saving that earned the credit shown next to it.
And the thing that does not change on purpose: if the number in the diff column is zero, so is the number in ours.