Skip to content

Billing and usage ​

The hosted service uses your personal Yi Labs wallet. Open Organization → Billing, then follow the Yi Labs account link to manage your plan, top-ups and invoices. Account help explains sign-in, credit eligibility and expiry.

Standalone deployments without ACCOUNT_URL have a separate organization billing guide.

Plans ​

Plans belong to a person, not an organization or a shared team. Each person has one workspace. Account subscriptions provide credit; the product still has per-run and resource safeguards.

PlanMonthly priceMonthly credit
Free$0$5
Pro$20$25
Max$100$125
Ultra$200$250
EnterpriseBy contractBy contract

Yearly subscriptions cost 20% less than twelve monthly payments; credit is issued monthly. Current prices and available payment actions are on Account Billing.

Credit and balance ​

Review reserves funds from your Yi Labs wallet, settles actual usage and releases unused reservations. Available balance includes the unspent funds reserved for Review; a reservation is not an extra charge. Billing in Account shows grant sources, expiry dates, product usage and the ledger.

Free credit is issued for each UTC calendar month. New-user credit is $20 once per mailbox and expires after 30 days. Free monthly and new-user credit cannot be used for Cloud. Paid-plan credit expires with its monthly slice; the Account page shows each grant's eligibility and expiry.

Subscribe ​

  1. Open Account Billing, directly or through Review's billing link.
  2. Choose Pro, Max or Ultra and the monthly or yearly interval.
  3. Complete the checkout offered there and return to Account to verify the plan and credit.

Review's hosted billing page does not sell a separate organization subscription.

Change plan ​

Use Account Billing to change your personal plan. Review the effective date and payment shown there before confirming. Plan management, credit and invoices stay in Account.

Cancel ​

Cancel renewal through Account Billing. The page shows the end of the paid period; cancelling renewal is not a refund request.

Top up ​

Choose Top up in Account Billing. Credit is added after payment confirmation. A top-up does not change your plan. If a payment is missing, contact support with its invoice or payment reference before paying again.

Redeem a credit code ​

Redeem supported codes in Account Billing. Historical Review PW-… codes cannot be redeemed in the hosted product: its local redemption API returns 409 managed_by_yi_labs. Ask admin@nice2.io about an unused historical code; it is not automatically converted into Account credit.

Monthly spending cap ​

With the Yi Labs wallet there is no monthly spending cap. The wallet funds usage; a monthly cap request here is refused with managed_by_yi_labs. Optional budgets you set on online-scoring rules remain available and are off by default.

What is billed ​

ItemHow it is counted
Sandbox timePer second from when the virtual machine is created until it is deleted, at least one minute each, at the class's price. Pre-runs and "Check once" count. A virtual machine that never reached the platform is not charged.
Judge modelThe platform's judge (the project's model, see Models, or the model an LLM scorer names): in cloud runs and chat AI tests, in scorers that ask a judge (an SDK experiment's platform scorers, "Try it", calibration), in the Playground's scorers and in online scoring: the model bill plus 30%.
Platform modelsCalls on a platform model through the AI Gateway and the Playground's prompt columns, a chat AI tested on a platform model, and the model calls of an agent in a repository or image project with no OPENAI_API_KEY in its secrets (your agent's model), at the same rate.
AssistantThe in-app assistant and the platform AI's other steps for a project (AI-written dataset rows and the issues it finds included), at the same rate.
StorageRun bundles and evidence kept on the platform, per GB per day.
Traces uploadedTraces from the SDK, OpenTelemetry and the Gateway, by volume. Listed on the usage page, not charged for now.

Your own model calls (your agent's with its own key, your app's through the Gateway with x-provider-key, an SDK Eval's task, a dev AI's) go to your own provider account, and a local test runner uses no sandbox time. The prices are in unit prices. What the platform runs again because of its own problem isn't charged; see the platform's problem isn't charged.

See your usage ​

"Organization" → "Usage" shows your workspace's Review usage one month at a time; pick another under "Month". Account shows the shared wallet and usage across connected products.

  • The month's total, and what it is made of: "Sandbox", "Judge model", "Platform model (AI under test)", "Platform AI", "Storage" and, once traces arrive, "Traces uploaded" (their size and number of spans; not charged for now), each with the amount used and its cost.
  • "Sandbox hours this month", against the limit an owner set in the organization policy, if any.
  • The available shared-wallet balance. Manage the plan and wallet in Account; there is no Review monthly spending cap in hosted mode.
  • "The platform's problem: not charged": what the platform ran again because of its own problem, by reason, with what it would have cost (see below). It isn't shown in a month without any.
  • "By project" and "By day".

The figures are an estimate of usage to date, in US dollars as billed; the invoice is what counts.

The platform's problem isn't charged ​

What the platform runs again because of its own problem is recorded as usual, but not charged (since 19 October 2026):

  • The platform's fill-in run after a sandbox was lost part way. None of it is charged: its sandbox time, its agent's calls to the platform model, its judging. It goes ahead even when the wallet's available credit is used up, held only to each run's own model and judge spending limits.
  • When the sandbox of a run or a pre-run is lost part way, its time after the last partial result it handed in. One lost before any attempt ran isn't charged at all, not even the minimum minute. One where attempts ran but no partial result ever came back is charged as usual: which part ran for nothing can't be told.
  • Re-runs by the platform caused by the platform model failing or the platform's own test backend failing: both tries' sandbox time, and the re-run's model calls and judging. Since 20 October 2026 the platform model failing includes the model the platform put in place of the one asked for refusing a setting of the request, whatever the agent then replied; such an attempt isn't judged either, so no judging is charged for it. Since 5 November 2026 it includes the agent giving up on the platform's model after waiting 30 s or more with no answer.
  • A retest ("Re-run these rows" on the experiment) of attempts that only the platform's model left untested (since 20 October 2026): none of it is charged. A retest that also takes attempts left untested for any other reason is billed as usual.
  • A judge call the second judge model made because the first was rate limited (since 22 October 2026): charged at the second judge's price, never more than the first one's for the same tokens. What would be over isn't charged. The same holds when the platform judges a run again after it.
  • A pre-run or check that failed only because of the platform's own part (since 22 October 2026), a run that was its own pre-run and failed so included: its model for the agent, its judge model, its tool proxy, or the test runner's own error. None of its sandbox time is charged; the pre-run page says "This sandbox time isn't charged." Its model calls are charged as they were made, as above.

Still charged: re-runs caused by an error in your own tools, data reset, database, keys or agent; a fault or an attack that never reached the agent; the agent unable to reach the model address on the sandbox itself; a job that "Lost contact" (its lease ran out); runs cancelled or stopped at a limit; a "Run the N that didn't run" you click; the first try's model calls before a re-run (charged as they were made).

The usage page's "The platform's problem: not charged" row gives what it would have cost this month, split by reason: "Filled in by the platform (a sandbox was lost)", "Re-runs by the platform", "A lost sandbox's time that left no result", "Retests of what the platform's model left untested", "The second judge's extra, used while the main judge was rate limited" and "Pre-runs and checks that failed only for the platform's own problem", with the hours where there is sandbox time. None of it is in the total, nor taken from the credit, the balance or the spending cap; its sandbox time doesn't count toward the month's sandbox hours, though it is in the "Sandbox" row's hours. On the job page of a platform fill-in run, the sandbox's cost reads "The platform's problem: not charged".

The billing page ​

Review displays your wallet balance and the link to manage it in Account. The product's Usage page shows its own usage; Account shows usage across connected products. Historical standalone balances are not automatically transferred.

Invoices and receipts ​

Open Account Billing for payment invoices, payment management and the ledger CSV. Product usage and a reservation are not separate card payments. For refunds, contact admin@nice2.io with the relevant invoice reference; see the account help.

Prepaid balance ​

Top-ups and eligible credit in Account fund new hosted usage. Historical standalone prepaid balances follow the standalone guide; contact support to resolve a balance mismatch before starting new paid work.

Spending limits and the estimate ​

With the Yi Labs wallet, the assistant, AI review, generating attacks and dev-AI connections have no daily spending allowance, and neither the platform's judge nor the AI under test's platform model has a per-run spending cap. Available wallet funds pay for them: judging and the platform model's answers go on as long as the wallet can fund them, and when it can't, the run stops with the usual low-balance message and what wasn't judged stays undecided. Each run retains its own step and model budget; sandbox concurrency, runtime and resource safeguards still apply. A failed or incomplete evaluation is not a pass.

The launch page shows the estimate and the limits for that run. An optional budget you explicitly set on an online-scoring rule still stops that rule when reached. See spending controls and organization policy.

When tests cannot start ​

If funds are insufficient, inspect available credit and reservations in Account Billing, then top up or wait for eligible credit to renew. A suspended account, unresolved debt or an unavailable Account service can also prevent new work; topping up does not resolve every refusal. Read the product's reason and contact support where needed.

Next steps ​

The legal texts on this site are drafts pending legal review and are not yet in force.