Skip to contentKumoDocs
Sections
On this page
Account and billing

Plans

Pro, Max and Ultra plans as the account's limit: buying, validity and renewal, expiry of what is left, model multipliers and the cache discount, which keys spend the limit, freezing, notices, and how a plan differs from Pay as you go.

View as Markdown

What a plan is

A plan is a token limit for a period, paid in advance. There are three plans: Pro, Max and Ultra. The bigger the plan, the cheaper each token in it, the longer its limit lasts and the larger its discount on reading from the cache.

The limit is spent in tokens rather than in money, and on its own it never draws on the wallet.

An account holds one limit. Every plan purchase adds its volume to that same limit — there is no choosing where it lands. Each purchase still keeps its own validity (see Validity).

The limit is spent by every key of the account that you have allowed to spend from it — and which keys those are, you decide key by key.

Pro, Max and Ultra

PlanHow it is boughtLimit validityDiscount on cache reads
Proa fixed volume at a fixed price1 month20 %
Maxa fixed volume at a fixed price2 months25 %
Ultraa volume of your choice: from the starting volume to the maximum, in fixed steps; the price grows with every step3 months30 %

Prices and volumes are published by the price list, so they are not repeated here: the current figures are on the pricing page and in the console, on the Plans screen. A plan's volume is stated on ×1 models — what that means is in How models spend the limit.

Need more than Ultra offers, or invoiced payment for a company — write to us, and the terms are set separately.

The limit and its gauge

What is left is shown as a percent of the limit: on the Limit card of the Billing screen, in the limit window in the console header, and on the spend screens as "% of limit". The console prints no token figure for the limit.

Next to the percent stands the plan's name — the highest of the plans whose limit you currently hold.

The gauge counts 100 % from your plan size: your largest plan purchase, or the sum of your purchases within 30 days if that is larger. The plan size never goes down, and when what is left after a purchase exceeds it, the gauge stands at 100 %.

When five days or fewer remain before the nearest remainder expires, an "expires … · … left" line appears under the percent. The nearest expiry is always visible in the Limit details window, in the Term row.

Buying a plan

A plan can be bought two ways:

  • on the Billing screen — with the Buy a plan button on the Limit card; the payment window opens on the Plan tab, where you pick Pro, Max or Ultra;
  • on the Plans screen — with the Buy Pro, Buy Max or Buy Ultra button on the plan's card. For a plan whose limit you already hold, the button reads Buy more. The Compare tab of the same screen puts the plans and Pay as you go side by side.

Ultra's volume is picked with the Ultra volume slider — only from the published steps, and the price is recomputed at every step. Pro and Max each have one volume.

A plan can be paid by card, through SBP, in crypto or from the balance. From the balance the plan is credited at once, with no payment page.

A plan payment is not refunded. When you buy a plan, the payment window always asks you to tick your acceptance of the public offer, and that tick is also your acceptance that a plan payment is not refunded. Without it the payment does not start. This applies to a plan paid from the balance too: the money spent on it stops being part of the wallet balance.

Until you go to the payment page, nothing has been charged. The limit grows once the payment is credited.

How a payment goes through → The price list The public offer

Validity, renewal and expiry

Every plan purchase has its own validity: Pro — 1 month, Max — 2 months, Ultra — 3 months from the moment it is credited.

What expires first is spent first. Of several live purchases, a request spends the one whose term ends soonest; with equal terms, the one bought earlier.

Any plan purchase renews what you already hold. Remainders of earlier purchases that would expire before the new one get the new purchase's term. A purchase never shortens a later term.

An example: Pro is bought on 1 March and lasts until 1 April. On 20 March you buy Max — it lasts until 20 May, and what is left of Pro now lasts until 20 May too. Had you held Ultra until 1 June, its term would stay as it was.

When a purchase's term ends, its unspent remainder expires: it is no longer spent and is not refunded. The expired remainder leaves the limit, and the percent on the gauge shows it.

Packages bought before plans were introduced never expire — see Packages bought earlier.

How models spend the limit

Every model has its own multiplier ×m — it is shown in the model catalogue, in the Multiplier column. A million tokens of a ×1 model spend a million of the limit, of a ×2 model two million, of a ×0.5 model half a million. That is why a plan's volume is stated on ×1 models: on a ×2 model it covers half as many tokens.

Cache reads cost the plan's limit less. Input tokens read from the cache spend the limit at ×m × (1 − d), where d is the plan's discount: Pro — 20 %, Max — 25 %, Ultra — 30 %. The discount in force is that of the plan whose limit the request spends — the one that expires first. Requests paid from the balance, and the remainder of packages bought before plans, get no cache discount.

Each model has one multiplier, printed exactly as the price list publishes it — in the price table, in the model catalogue and on the model's page. Its hover card says how many of the model's tokens a million package tokens buy. The cache discount is not a second multiplier: it applies only to input read from the cache, as above, and the limit is charged by the actual tokens of each request.

What counts as spend

Different models have different prices, and a request to a more expensive model takes more of the limit than the same request to a cheaper one.

Input is everything sent to the model — the system prompt, the message history, the tool definitions, the attached files, and anything read from or written to the cache. Output is everything the model generated, reasoning tokens included, even when you do not see them in the answer. Both count toward what a request spends from the limit.

The token counts a request went to the model with are visible on the activity screen and in the expanded call card. How much that request took from the limit, the console shows as "% of limit".

The practical upshot: a long conversation history is input, paid for again on every request. In agentic work it is usually the history, not the model's answers, that makes up most of the spend. When the history repeats from request to request, part of it is read from the cache — and for a plan's limit that is cheaper.

Which keys spend the limit

Each key chooses for itself what it spends from: the limit or the balance. The choice is not final — change it at any time in the Choose how it is charged window, opened by pressing the key's cell in the Billing column of the keys screen (the Spends from row: Limit or Balance).

For a key that spends from the limit, the same window sets what to do once the limit runs out (the After the limit row):

Stop
The key refuses requests until you buy a plan.
Balance
The key carries on from the balance, at wallet prices.

Until you choose, a key spends from the limit when the account has one and stops when the limit runs out; without a limit the key spends from the balance.

Each key can also carry its own ceilings on the limit (Key ceilings on the limit): how much it may spend from the limit per day, per month and ever. An empty field means no ceiling. That way one hungry agent cannot eat the limit the other keys rely on.

A project changes none of this: it is a label over keys, not a source of money.

Keys → How the wallet counts →

Freezing

The limit can be frozen with the Limit switch (Active / Frozen) in the Limit details window — opened by the button in the corner of the Limit card. While it is frozen, keys that spend from the limit refuse requests with `403` package_frozen — and they do not fall back to the balance, even when "After the limit — Balance" is chosen for them.

The same switch unfreezes it. Keys that spend from the balance are not affected by a freeze.

A freeze does not stop the clock: plan purchases whose term ends while the limit is frozen expire as usual.

This freeze is yours. Do not confuse it with a freeze by Kumo: if the limit has seen no purchase for about three months, Kumo may freeze its unspent remainder. That is Kumo's decision, not a timer. The card then reads "The remainder is frozen — buy a plan to unfreeze it", keys on the limit refuse requests with the same 403 package_frozen, and the switch does not lift that freeze — the next plan purchase does.

Notices

The console reminds you about the limit in its notifications window twice: when less than 20 % of the limit is left, and when the limit is used up. Buy a plan, rise back above the threshold and drain it again, and the reminder comes again.

When the limit runs out

What happens to a request is decided by the key's After the limit choice.

  • Stop — the request is refused with `402` and the code insufficient_package_balance, a code separate from the wallet's, because the cure is different: buy a plan rather than top up a balance.
  • Balance — the request goes through and is paid from the balance at wallet prices. If the balance is empty too, the refusal is the wallet's: 402 insufficient_wallet_balance.

A request started while something was left may, on its way, spend more than the limit still held. For a key on Stop, that overrun is not taken from the balance — Kumo absorbs it.

Do not confuse it with scope: a call naming a model outside the key's whitelist is a 403, and it is cured by a different key rather than by a purchase.

Every gateway status →

Packages bought earlier

Packages bought before plans were introduced keep working and never expire. What is left of them belongs to the same account limit and is spent after your live plans: whatever has a term is spent first. That remainder gets no discount on cache reads.

The price of such a package, its unspent remainder included, is not refunded either. The exact terms are in the public offer.

A plan's price in roubles

A plan is priced in dollars: the platform buys the capacity from its suppliers in dollars, and the dollar price does not move with the rouble.

The rouble figure on the storefront is that same dollar restated at the accounting rate of the Bank of Russia. It is there to orient you.

What you are charged is a different figure: it is computed at the payment's rate, pinned at the instant the payment is created, and raised up to the next ten roubles. The purchase window shows it before you go to the payment provider — that is the figure the bill carries.

So the roubles on the storefront and the roubles in the purchase window may differ, in either direction: the two rates are different and both move from day to day. The plan's dollar price is the same one throughout.

The mark tells the two apart: a rouble figure with an ≈ is a restatement, one without it is the figure you will be charged. That is why every rouble on the storefront carries the mark, while in the console only the rows a price is built from do — the price itself and the amount to pay carry none.

About both rates →

Plan or Pay as you go

Pay as you go is a top-up of the wallet balance: from $3, by any amount above that. The balance never expires, and calls made from it are charged at the price list's wallet prices, with no plan discount.

Pay as you goPlan
What is showndollars on the balance% of the limit
Funded bya balance top-up from $3buying Pro, Max or Ultra
Validitynever expireseach purchase its own: 1, 2 or 3 months, renewed by the next purchase
Discount on cache readsnone20 %, 25 % or 30 %
Charging rulea price per unit of every dimensionthe model's multiplier, against the limit
Refundthe unspent remainder — under the public offer's rulesnot refunded
Refusal when short402 insufficient_wallet_balance402 insufficient_package_balance

Running both is legitimate and ordinary: some keys spend from the limit, some from the balance, and a key on the limit can carry on from the balance once the limit runs out. Each key's spend is visible on its own — from the limit and from the balance apart. What cannot be done is adding the two into one number: a sum of dollars and limit tokens has no unit.

The price list Where to watch spend →