Pay per token, or take a weekly allowance.

A token is a short piece of text, on average about three quarters of an English word. Models count the tokens they read (input: your prompt, files, chat history and tool definitions) and the tokens they write (output: the reply and any reasoning). Every rate on this page is per one million tokens.

Input tokens times the input rate, plus output tokens times the output rate. Where a model has a cached-input rate, input served from the cache bills at that lower rate. For example, 20,000 input and 2,000 output tokens on GPT-5.6 Sol cost $0.176. Each request is rounded to the nearest $0.000002, the smallest amount the gateway bills.

Purchased credit does not expire while your account is open and in good standing, as set out in the Terms of Service. Plan allowances work differently: unused allowance does not carry over to the next refill.

New accounts start at $0, with no sign-up credit. Models, free models and promotions become available once the account has paid $3 or more, by a top-up or the Pro trial. Each top-up is at least $3, and the whole amount is yours to spend on any model.

Requests return an insufficient-balance error until you add credit. A reply that is already running finishes, so the balance can end slightly below $0; the next top-up covers the difference. There are no free models at the moment.

Pay $3 once by card and get the first month of Pro: 3.4M uncached DeepSeek V4 Flash input tokens a week, refilled every Monday. It is limited to one per account and does not renew; after the month you can choose any plan or stay on pay as you go. Like the plans, the trial is opening soon.

A plan's allowance is a weekly amount of usage at the plan rates, measured in DeepSeek V4 Flash input tokens. The uncached number is what it always pays for. Input the provider serves from its cache uses 1/50 as much allowance, so with about 76% cache hits the same allowance pays for the "up to" number (Pro 3.4M uncached, up to 13M; Plus 6.4M uncached, up to 25M; Max 21.2M uncached, up to 81M). Output tokens count 4 times, because they cost 4 times as much. On a model with higher rates the same allowance pays for fewer tokens.

Every model. Each request uses the allowance at the published rate, less the plan's discount (Plus 2%, Max 4%), so a model with higher rates uses it up faster. The token figures are measured on DeepSeek V4 Flash. When the allowance runs out, the wallet pays at the same plan rates.

There are two kinds. A model promotion lowers that model's rate for everyone until its end date. A plan discount (Plus 2%, Max 4%) lowers the rates of every model while the plan is active, from the allowance and from the wallet, on top of any model promotion. The allowance token figures already include it.

Plans are opening soon. A plan is paid for one month at a time and does not renew automatically, so there is nothing to cancel: it simply ends at the end of its month, and pay as you go continues from your wallet.

Online checkout is not open yet. On the billing page of the console, choose a credit pack and email us with your username. We reply with payment details and add the credit to your account once the payment is confirmed.

Not automatically yet. The billing page lists every credit added to your account with its date and amount. If you need a receipt for a pack, ask for one when you buy it by email.

Yes. Create as many API keys as you need under one account; every key draws from the same balance, and each key can have its own spending limit or be revoked at any time.

Unused purchased credit can be refunded within the window set out in the Refund Policy. Credit already spent cannot be refunded. Email support@relaykit.dev from your account address.