Input and output are different prices
A model's input price applies to prompt tokens; its output price applies to generated tokens. Catalog prices are expressed in USD per one million tokens. To estimate a basic text request, divide input tokens by one million and multiply by the input rate, then add output tokens divided by one million multiplied by the output rate.
For an illustrative calculation only, 2,000 input tokens at $1 per million and 500 output tokens at $4 per million total $0.004. These example rates are not a Kearnix quote. Use the current model detail page for actual prices.
Compare the same workload
The lowest input rate is not necessarily the lowest total cost. A model with a higher output rate can cost more for long answers even when its prompts are inexpensive. Test with representative prompts and response lengths, then compare the cost of useful results rather than the advertised rate in isolation.
Context window is a capacity limit, not the number of tokens that every request uses. Long conversation histories, retrieved documents and tool definitions increase input size. Output limits are upper bounds; actual consumption depends on the response and the model's billing rules.
Check caching and other billing units
A separate cached-input price can apply only when the relevant model and request actually qualify for caching. A missing cached price does not establish a zero cost. Some models distinguish cache writes or use other billing units for images, audio or additional features. Read the model detail rather than applying the simple text formula to every endpoint.
The Kearnix catalog displays customer prices. You do not need to add a separate wholesale markup to those displayed rates. Account usage is the place to reconcile the requests you actually made with consumed credit.
Start small and measure
Use prepaid credit and a small evaluation batch before expanding traffic. Record the model ID, request count, input tokens, output tokens and total spend. Review unusually long prompts or responses and set application-side limits that match your use case.
Create an account when you are ready to test, then choose a model from the live pricing page. Revisit your estimates when you change models, context length or output patterns; static price comparisons go out of date as catalogs change.