typed/recipes

What does Jev cost?

Updated 24 Sep 2026

Jev costs $0.042 per million input tokens, and output is free. TypeSafe lists no per-question fee, so your bill tracks how much text you send. Laya, the open-weight alternative, has no per-call fee; you pay for the hardware you run it on.

How do I estimate my bill?

Multiply the input tokens per call by your call volume, then by $0.042 per million. A 1,000-token support ticket costs about $0.000042 per call, so a million tickets cost about $42 in Jev charges.

If you ask several questions about the same text, send them in one call instead of several, then confirm in your usage logs how the tokens were counted. Gateways can count differently from TypeSafe’s own API.

What have builders reported spending?

These are figures authors published in launch week. We list them because they are specific, not because we have reproduced them.

JobReported costReported byTested by us
Classify 500 emails3.5 centsRiley Brown, X postNot yet
Classify 1,018 research papers$0.08 for Jev, plus $3.99 for LLM summariesHassan, 1kpapers.comNot yet
1,709 editorial judgmentsUnder $0.01EveryNot yet
Screen 100 emails for fraud, uncertain ones sent to an LLMAbout $0.003 for Jev, about $0.07 in totalHassan, X postNot yet

The fraud pipeline is the useful lesson. The LLM fallback cost over 20 times more than the Jev calls. In a cascade, your fallback rate decides the bill, which matches what we saw in the fallback benchmark entry.

Is Jev really 444× cheaper than an LLM?

That figure comes from workflows TypeSafe’s own team built, and TypeSafe says the gains likely sit at the high end. The multiplier depends on which model you compare against and how much text each call carries. See our verdict on the claim.

Where can I buy Jev?

Directly from TypeSafe AI through its API (documentation). Vercel AI Gateway, Cloudflare AI Gateway and OpenRouter (in beta) have also announced Jev access. Check each provider’s model page for its price and token counting before you commit.

Is Laya cheaper?

Per call, yes: Laya is free under Apache 2.0 and runs on your own GPU or CPU. The base checkpoint is near chance on typed decisions without training (0.362 against a 0.318 random baseline), so budget for labeled examples and fine-tuning. See the Laya page and the side-by-side comparison.

Sources and test status

Reviewed 24 September 2026. The price is from TypeSafe’s launch post and documentation. Reported bills are from the authors’ public posts. We have not yet measured Jev’s cost on our own workload.

Continue learning