Jev Pricing

01 / FIRST-HAND MEASUREMENTS

What our real Jev test calls used

Our Phase 0 validation sent three real requests to typesafe-ai/jev through Vercel AI Gateway on September 18, 2026. These are three specific test requests, not a general average for Jev. The input costs below are calculated at the current published USD rate.

Recorded during our initial technical validation; these are not the later production examples in the API guide. Token counts come from the saved SDK responses. This is a small functional sample, not a cost or performance benchmark.

02 / WHAT YOU PAY FOR

Input length sets the model cost

Count the complete request, not just the customer’s message. Input tokens can include the state being evaluated, question instructions, Choice labels and descriptions, Score rubric definitions, and each additional question. One user click does not imply a fixed number of tokens—or even a single API request.

03 / WORKED EXAMPLES

From ten thousand calls to ten million

Illustrative workloads at $0.042 per million input tokens, equivalent to $42.00 per billion. These are model input costs for the total call counts shown, with no assumed time period.

The calculator and every cost example on this page use the same rate and calculation. If retries create more calls, include them in your request count.

04 / THE REST OF YOUR BUDGET

What the calculator does not include

This estimates Jev model charges. It does not price an entire application or guarantee your account’s bill.

• Vercel hosting and other AI models
• Databases, storage and observability
• Retries beyond the entered request count
• Application engineering and human review
• Taxes, optional Gateway features and account-specific terms

Budget for useful decisions: retries, human review and downstream errors can exceed the model bill. Track input tokens, API cost, decision quality, review rate, retry rate and latency. Confirm your provider account terms; Jev Playground does not resell API access.

05 / REPLACE ASSUMPTIONS WITH USAGE

Measure your own requests

After a successful server-side evaluation, AI SDK 7.0.105 exposes result.usage.inputTokens and result.usage.outputTokens. Either count can be undefined. Treat missing usage as unknown; do not silently record it as zero.

// After a successful evaluate(...) call in your server code:
const inputTokens = result.usage.inputTokens;
const outputTokens = result.usage.outputTokens;

if (inputTokens !== undefined) {
  console.log({ inputTokens, outputTokens });
} else {
  console.log("Input token usage was not reported.");
}

Sample short and long inputs from each workflow. Use total measured input tokens divided by measured requests to update the calculator’s average, and track requests with missing usage separately. Keep raw customer state out of usage logs.