Quick answer
As checked on September 20, 2026, TypeSafe lists Jev input at $0.042 per million tokens, equivalent to $42 per billion. Output tokens are listed as free. That is a token rate, not a promise of free API access or a fixed cost for every decision.
Estimate cost from measured input tokens
Use the input_tokens value reported by your own requests. For planning, multiply requests by average input tokens, divide by one million, then multiply by $0.042. The examples below assume 1,000 input tokens per request and the published rate; they are arithmetic examples, not measured bills.
| Requests | Assumed input tokens | Estimated input cost (USD) |
|---|---|---|
| 1,000 | 1 million | $0.042 |
| 10,000 | 10 million | $0.42 |
| 100,000 | 100 million | $4.20 |
| 1,000,000 | 1 billion | $42.00 |
Budget for the workflow, not just one call
A user action may cause several model calls. A routing step, a second verification and a retry all affect the total. Measure the number of calls per completed task instead of assuming that one visitor produces one request.
Model charges are only one line of a production budget. Your backend, data preparation, logging and human review can cost more than inference. Track them separately so that a low model rate does not hide a more expensive operating process.
- Measure the average and high-percentile input size from representative tasks.
- Include unsuccessful attempts and retries when estimating call volume; confirm actual billing treatment in your account.
- Compare cost per correctly completed task and review workload, not only cost per token.
Free output tokens do not mean unlimited free use
The published output-token price does not remove input charges. This guide does not verify any permanent free tier, signup-credit allowance or minimum account spend. Check the official console before budgeting around promotional terms.
Likewise, a price table does not guarantee that an account is already approved. Access, account limits and billing settings need to be checked separately. Contact TypeSafe for contractual or enterprise terms rather than extrapolating them from a public example.
Plan for limits and changing model versions
Request limits and context limits are different constraints: the first controls throughput, while the second controls what fits into one call. The official models page warns that rate limits can change during the rollout, so copy current values from that page when configuring production capacity.
For repeatable cost comparisons, record the resolved model version, input usage and date. A moving alias such as jev-latest can change over time. Repeat your sample when prompts, input size or model versions change instead of treating this page’s example as a permanent quote.
How to compare Jev with another model fairly
Use the same input records and require equivalent outputs. If one model supplies probabilities and another only returns a label, the two results serve different workflows. Decide which output you actually need before calculating savings.
Count errors that need manual correction. Saving a small amount on inference is not useful if incorrect routing consumes more staff time. Start with a replay of a fixed sample and compare total cost, latency, decision quality and the fraction sent to review.
Common questions
What would 100,000 calls with 1,000 input tokens each cost?
At the checked rate, the input-token estimate is $4.20. It excludes other services and assumes the stated token count and rate.
Is $42 charged per million tokens?
No. The published figures are $42 per billion input tokens or $0.042 per million.
Is this a guaranteed quote?
No. It is a dated explanation of published pricing. Current official pricing and your account terms take precedence.
Sources & verification
Examples and explanations are editorial guidance. Current official documentation and account terms take precedence.