Jev Pricing

Jev pricing is easy to misquote. This page only repeats figures that appear in a source, and it dates them.

Last verified: 2026-10-16. Updated: 2026-10-16.

Published public price

The same rates feed the Jev Cost Calculator. The calculator uses the same published rate.

Model aliases (same price, different weights)

Aliases can move when TypeSafe ships. Log response.model and pin a versioned ID after you tune thresholds.

What you actually pay for

You pay for tokens in state plus tokens in questions. Official docs emphasize that extra questions are cheap relative to repeating the state. A cookbook on TypeSafe's site reports large savings from batching many questions in one call. That speed/cost multiple is a vendor cookbook result, not an independent test.

Rate limits (not the same as price)

When last verified, the public models page no longer lists fixed token or request caps. Limits are per organization and shown in the TypeSafe developer console. Crossing a limit returns 429. Official SDKs retry with backoff.

Enterprise and custom limits: official docs point to sales@typesafe.ai.

Long context (same token price, different risk)

Optional long context raises both limits to 262k tokens (state plus all questions, and state plus any one question) when you send:

TypeSafe-Danger-Zone-Long-Context: true

Official docs say long context costs the same per token but is slower and less accurate than staying within the default 64k / 32k limits. See the Jev API page for the full note.

Other cost paths

A third-party article reported the same $0.042 / million input price on Vercel AI Gateway under typesafe-ai/jev. We have not independently confirmed Gateway billing in this project. Check Vercel's current table if that is your path.

Worked estimate

100,000 requests/month, 400 input tokens of state, 3 questions ≈ 40 tokens each:

• tokens / request ≈ 520
• monthly tokens ≈ 52,000,000
• monthly cost ≈ 52 × $0.042 = $2.18

This uses the published input rate and an assumed 40 tokens per question. Real usage comes from the usage object on each response.

When Jev pricing looks attractive

High-volume classification, routing, and filtering where an LLM would burn output tokens to emit JSON.

When it does not matter

If you make a few hundred calls a day, engineering time dominates. If you need generated text anyway, Jev does not replace that bill.

Common mistakes

• Hardcoding $0.042 in five files. Use one config.
• Forgetting that aliases can change behavior without changing the price row.
• Comparing Jev's input price to an LLM's input price and ignoring that the LLM also bills output.
• Turning on long context for every request instead of only payloads that need it.

See the API for usage fields and the changelog when TypeSafe updates the models page.