Jev pricing explained

TypeSafe's launch material lists Jev at $0.042 per million input tokens and says output is free. That is equivalent to $42 per billion input tokens.

The number is unusually low compared with generative language models because Jev does not generate a prose response token by token. It returns the typed decisions declared by the request.

The published price is a provider snapshot

The $0.042 figure appears in TypeSafe's launch post and on OpenRouter's TypeSafe listing. Vercel also publishes current Jev pricing on its AI Gateway model page.

Provider pages can change independently. Before estimating a production bill, confirm:

• the exact model identifier;
• input price and any free-tier terms;
• whether the provider adds markup or credits;
• how usage is measured;
• retention and training settings;
• whether the route is generally available or early access.

Why “output free” does not mean the whole system is free

Jev's returned decision payload is not billed like generated prose on the published direct pricing. Your total application cost can still include:

• gateway or platform charges;
• retries and failed requests;
• logs, storage, queues, and observability;
• a fallback language model for uncertain cases;
• human review for decisions below your confidence threshold;
• evaluation work needed before automation is safe.

For many teams, the expensive part is not inference. It is deciding when the model is reliable enough to act without a person.

A useful cost model

Estimate cost per completed workflow, not cost per token. Count every Jev call, fallback call, retry, and manual review triggered by the workflow.

Then measure two rates on representative labeled data:

1. Coverage — the share of cases your threshold allows the system to automate.
2. Error cost — the consequence of a wrong automated decision.

A lower threshold may automate more work while increasing mistakes. A higher threshold may make inference cheap but keep human-review costs high. The right threshold is a product and risk decision, not a model default.

Treat benchmark savings as vendor-reported

TypeSafe reports that Jev was up to 193.6× faster and 444.6× cheaper on its workflow evaluations. The company also explains why those figures may be near the high end of real-world gains.

Use the published evaluations to understand the workload and methodology. Do not copy the maximum multiplier into a business case without running your own representative evaluation.

For access options and current provider links, see Jev API Access.