Get started

Pricing and billing

Prices per million tokens in rupees, prepaid credit, and what happens when your balance runs low.

How you're charged

There is no subscription. You pay for the tokens each request uses, at the model's prices in the catalogue:

cost = (prompt_tokens − cached_tokens) × input price
     + cached_tokens × cached-input price
     + completion_tokens × output price

When part of your prompt repeats an earlier request, for example an agent sending its conversation again, those tokens may be read from cache and are billed at the lower cached-input rate. Each model's rates are on the Models page. cached_tokens is part of prompt_tokens, never an extra amount, and a model with no cached-input rate bills its cached tokens at the input price.

Prices are per million tokens, in rupees, and the cost comes out of your organisation's prepaid balance. Streaming and non-streaming requests cost the same. A request that fails is not charged; a stream that ends early is charged for the output it delivered.

Every response reports its tokens in usage, and every request's cost is in your request logs.

Credit

A new organisation starts with Rs 1,000 of credit, enough to integrate and evaluate. To add credit, talk to us and we'll add it to your balance. Self-service top-up is not available yet.

When your balance runs low

Before a request runs, Paranine checks that your balance, and the key's credit cap if it has one, can cover the request's maximum cost: the prompt plus max_tokens at the model's output price. If it can't, the request answers 402 (insufficient_funds) and is not charged.

So a request with a large max_tokens can be refused while smaller ones still succeed. Lower max_tokens, raise the key's cap, or add credit. New credit counts from the next request.

Capping what a key can spend

Every key spends from your organisation's balance. Give a key a credit cap to limit what that key alone can spend over its lifetime: useful for experiments, contractors and CI. See API keys.

Seeing what you spend

Request logs: the cost of every request, in rupees.
Each key's page: that key's spend over the last 30 days.
In your code: stream with X-P9-Debug-Metrics: true to get a request's cost as its stream ends; see Streaming.

Dedicated capacity

For reserved hardware, an SLA or invoiced billing instead of prepaid credit, see dedicated inference.