Jev pricing, explained

The published rate

Published model rates. Check the official source for current access and billing terms.

Price checked Sep 17, 2026Verify official pricing↗

The official models page is the source for the rate shown here. This is a dated snapshot of public pricing, not a quote for your account or a promise that the rate will stay unchanged.

What to put in the calculator

Use the number of API calls, not the number of documents or people, and the average input tokens per call. A document processed in three separate requests creates three calls. A single request containing several questions is still one call, with a larger input.

The calculation is:

estimated input cost =
  API calls × average input tokens per call
  ÷ 1,000,000 × published price per million input tokens

The calculator on this page runs in your browser and does not call Jev. It does not upload documents or measure their token counts.

A useful budget starts with measured inputs

For an initial estimate, use a deliberately simple workload, such as 10,000 calls averaging 1,000 input tokens each. Then replace that assumption with observations from your own integration.

The API reports usage.input_tokens. Use that field to examine representative short, typical, and long requests. Looking only at the customer message can underestimate the request: your question instructions, options, and other supplied context also matter. TypeSafe specifically notes that extra Choice questions and options add tokens. Choice documentation

Track separate call paths when their inputs differ substantially. A short routing decision and a long document review should not share an average merely because they use the same model.

Budget for the whole path

Suppose each document takes a first screening call, and some documents need a second review. Count both paths. If a retry resends the same document, it is another attempt; do not equate one user click with one request.

Keep a simple workload record: which route ran, how many calls it used, input-token usage, and whether a person had to intervene. Separate a deliberately replayed documentation example from a live request when reading your logs. A cached example does not establish today’s account charges or performance.

What this estimate leaves out

• Repeated calls, retries, and changes in average input length.
• Text extraction, retrieval, storage, hosting, and any additional model calls.
• Time spent reviewing uncertain results or correcting wrong decisions.
• Any taxes, account-specific commercial terms, or future price changes.

These are budgeting considerations, not assertions about additional TypeSafe charges. Check your account and current official terms for the actual bill.

Does a low API cost make the whole workflow cheap?

It helps only if the decisions are useful. When comparing approaches, measure the cost of getting an acceptable outcome, including review and recovery work. A lower-priced call that sends more tickets to the wrong team may cost more to operate.

A practical experiment is to run the same allowed examples through your current process and a Jev-assisted process, then compare decision quality, review rate, latency, and total calls. Our use-case guide offers a small starting workflow.

Pricing questions

Is the output-token field an extra charge?

The API can report output-token usage while the public model page lists output as free. Usage measurement and the current charging policy are different things. Refer to the published rate above and your account terms. Model pricing

Does adding more questions make a request free?

No. Questions can share one state and one request, but additional question content still affects input usage. Batch independent questions when that suits the workflow, then inspect actual usage rather than assuming unlimited work for a fixed price.

Are rate limits fixed?

TypeSafe says its early-access limits can change. Consult the models page for the current limits and use the integration guide for response handling. This calculator estimates spend, not available throughput.