Pricing and credits

One balance per workspace. One credit is one US dollar. There is no plan to buy: you add credits when you want them and pay for what you run. This page is the one place the money numbers live. Other pages link here rather than repeat them.

​

What a token costs

The two products are priced differently, because one is a model we operate and the other is a GPU you rent.

​

Model APIs

Priced per model, not by size band. When rates are published for a model, its page in the Model Library lists the input, cached input, and output rates per one million tokens. If that published price set has no separate cached input rate, cached input bills at the standard input rate, and the page says so. An unpublished fact is left absent rather than filled with a placeholder. Model APIs work on every account, whatever you have spent.

​

Your own deployed endpoints

Priced per million tokens by model size, input and output counted separately. Always-on endpoints pay less per token because they also pay an hourly fee to hold a GPU warm. That fee depends on the GPU class, the Deploy tab projects it for your exact configuration before you commit, and if your balance cannot cover the next hour the endpoint is stopped rather than served for free. Your per-token rate does not change with the GPU; GPU choice changes latency and throughput.

​

Optimization runs

Every run is quoted before it starts. The price follows the size of the models and the techniques in the plan, not how much you have spent. There are no included free runs and no monthly quota. The estimate you accepted is a ceiling. The final charge is never more than the number on the card you accepted, even if the run cost more to produce. A plan that changes enough to cost more is re-quoted, so a higher price is always a new decision, never a surprise on the receipt. A run that fails or that you cancel releases the whole hold. If the hold is larger than your available balance, nothing starts and no GPU time is spent: add funds, or select Reduce plan cost and let the agent revise the plan for a new quote.

​

Capability milestones

Capabilities unlock at lifetime credit purchases, automatically. You do not buy, choose, or renew a milestone. Your current limits are shown in Settings > Billing. RunInfra recalculates the milestone when a paid purchase settles; allow up to a minute for it to apply everywhere. The purchase that moves you up names the milestone you reached and the one you left, reading the same ladder that enforces your limits, so it never announces capacity you have not been granted. Before your first purchase a workspace can create an API key and call the Model APIs at 60 requests per minute, use the agent to build and plan up to 3 pipelines, run 100 playground requests a day, and run optimizations once the balance covers the quote.

​

Adding funds

• One-time top-ups from $10 to$10,000 at Settings > Cost. Pick an amount and RunInfra hands you to Stripe’s secure checkout, which collects your card and billing address, offers an optional VAT or tax ID field for business purchases, and shows the exact charge before you confirm. Every purchase issues a Stripe invoice carrying those details. Presets are $10,$25, $50,$100, $250, and$500, and you can enter any whole amount in range.
• Auto recharge is off until you turn it on at Settings > Cost. Enabling it runs one top-up through Stripe checkout with card saving on; after that, when your balance falls below your threshold, RunInfra charges the saved card, up to a monthly ceiling you set.
• Billing details live at Settings > Cost: set the company name, billing address, and tax IDs that appear on your invoices, manage saved cards, and browse every finalized invoice. Managed by Stripe.

A declined card stops auto recharge and flags it rather than retrying against a card that will not work. Credit is added when the payment confirms, never before, and an automatic recharge counts toward lifetime purchases exactly like a manual top-up.

​

Spending controls

Your balance is the real ceiling. Nothing that costs money starts without a balance that covers it. A workspace owner can also set a monthly spend limit on new optimization runs and new deployments at Settings > Billing. A soft limit emails you; a hard limit also refuses to start new runs and deployments. Email goes out as spend crosses 50%, 75%, 90%, and 100%. A spend limit is deliberately not a kill switch on your traffic: endpoints that are already deployed keep serving past it, and it does not govern auto recharge. To bound automatic charges, use the auto recharge monthly ceiling.

​

At zero balance

Nothing is faked, estimated, or quietly run on credit.

• New optimization runs, agent turns, and new deploys are refused before any GPU work starts, and the refusal states your balance and the amount required.
• Calls to the Model APIs and to your deployed endpoints return 402 Payment Required without reaching a GPU. See Errors.
• Always-on endpoints are stopped when the balance cannot cover the next reserved hour. They are not deleted, and they are not served for free.
• If a settlement drives the balance below zero, the workspace pauses new billable work. Your next top-up offsets the negative amount first.
• A screen that cannot read your balance says so and offers a retry. It never renders a zero it did not read.

​

Details you may need