Enter Token
Prompts sent to the model, system prompts, and conversation history are billed at the per-input rate.
Billing Rules
Billing is based on tokens, and these are the only rules.
There are no tiered pricing plans to pore over, nor are there any hidden surcharges buried in the fine print. Billing is based on just two factors: how many tokens this request used, and the unit price at the time of usage.
Billing Unit
Four Types of Tokens, Four Unit Prices
Within the same request, input, cache hits, output, and inference tokens are metered and billed separately. The billing details can be reconciled item by item, and different unit prices are not combined into a single total.
Cache Input Token
For input portions that hit the prompt cache, billing is based on the lower cache hit rate.
Output Token
Model-generated responses are billed based on the unit price per output, with the same pricing structure applied to both streaming and non-streaming formats.
Inference Token
Thought tokens generated by inference-based models are billed separately at the per-inference rate and are not included in the output price.
Prices are uniformly listed in USD per 1 million tokens. The backend records transactions as integers in micro-USD (1 USD = 1,000,000 micro-USD), which are converted to decimal USD values only when displayed on the page—ensuring that invoices do not accumulate discrepancies due to floating-point errors.
Billing Formula
One formula to calculate a single request
The total cost for this request is calculated by multiplying the usage of each token type by the unit price recorded for that request and then summing the results. The unit price is taken from the price list, and the usage is based on the gateway’s actual statistics; neither is estimated by the front end.
Cost = Input Tokens × Input Unit Price + Cache Hit Tokens × Cache Unit Price + Output Tokens × Output Unit Price + Inference Tokens × Inference Unit Price
The unit of price is USD per 1 million tokens; calculations are performed using integer micro-USD values, which are converted to decimal USD values only when displayed.
Live Example · The model with the lowest input price in the current directory
GPT-5.6 Luna Input 0.012 USD/1M × 1M + Output 0.072 USD/1M × 1M = 0.084 USD
This is a hypothetical request involving “1 million tokens each for input and output,” used to demonstrate the conversion method; actual billing is calculated based on the actual number of tokens per request.
Six Rules
Every penny on the bill corresponds to one of the items below.
These rules apply equally to both console billing and usage data returned by the API; there is no difference between what is displayed on the page and what is shown in the backend.
Pay-as-you-go: You're charged based on how much you use.
Each request is billed based on actual token usage: input, cache hits, output, and inference are metered separately. There are no monthly fees, no minimum spending requirements, and no idle fees.
The credit limit is valid indefinitely.
Your prepaid balance will not be reset to zero, will not expire on a monthly basis, and will not expire due to prolonged inactivity; every deduction is recorded in the ledger and can be verified line by line in the console billing section.
Price Snapshots, Historical Bills Remain Unaltered
When usage occurs, enter the current sales price in the corresponding usage record. Subsequent price changes will only affect new requests; invoices that have already been generated will always be calculated based on the price in effect at the time they were issued.
Failed requests are not billed
Requests that do not generate upstream output—such as those resulting in parameter errors (400), authentication failures (401), insufficient balance (402), unauthorized access (403), or rate limiting (429)—are not charged.
Streaming interruptions are billed based on the portion already transmitted.
SSE has begun to trigger upstream interruptions after output; it charges only for the input/output tokens that are actually generated and returns an `upstream_stream_error`, ensuring there is no double billing.
Payments and Invoices
We accept WeChat Pay, Alipay, UnionPay credit cards, and corporate bank transfers. After making a purchase, you can request an invoice yourself on the Finance page of the control panel; contracts and corporate banking procedures are handled through our corporate partnership.
Billing FAQ
Billing FAQs
Use the same set of answers as the solution selection page to avoid inconsistencies between the two. Write the answers directly in the HTML so they can be read even without executing JavaScript.
How does Admesh.ai charge for its services?
Pure Pay-as-You-Go: Billed based on the actual number of tokens used per request, with no monthly fees, no mandatory subscriptions, and no minimum spending requirement. Pricing is in U.S. dollars per million tokens, and invoices are itemized by individual request.
Do top-up credits expire?
No. The balance you top up remains valid indefinitely; it is not reset monthly, nor does it expire due to prolonged inactivity.
How are input, output, and cache tokens calculated, respectively?
Input tokens are billed at the input unit price, and output tokens are billed at the output unit price; input tokens that hit the prompt cache are billed at the lower cache-hit unit price. These three items are listed separately in the billing details for the same request.
What payment methods are accepted? Can you issue an invoice?
We accept WeChat Pay, Alipay, UnionPay credit cards, and corporate bank transfers. After making a purchase, you can request an invoice yourself on the Finance page of the control panel.
Will the price change?
We may adjust our prices in tandem with changes made by upstream vendors, and will announce such changes in advance via the console and in the release notes. Usage that has already occurred will be billed based on the price snapshot at that time, and historical invoices will not be altered due to price changes.
Will I be charged if the call fails?
No. Requests that do not generate upstream output—such as those resulting in parameter errors (400), authentication failures (401), insufficient balance (402), or rate limiting (429)—are not billed; however, the portion of the output already generated in the streaming response will be billed accordingly.
Now that you've read the rules, the next step is to choose a plan.
Low-traffic projects start directly with the trial plan, and top-up amounts are valid indefinitely; corporate contracts, invoices, and dedicated channels are handled through corporate partnerships.