Is Jev Free? Pricing, Access, and What Builders Actually Pay

Short answer: access and pricing are TypeSafe AI's to set, they change at launch-week speed, and docs.typesafe.ai is the only source that counts. This page exists because the more useful question has a stable answer: whatever the rate card says this week, what do real workloads cost?

Per the receipts builders publish alongside their projects, the answer is "almost nothing, but not nothing": 500 emails classified for about 3.5 cents (build), roughly 26,000 judge verdicts for $0.13 (build), browser tasks around a tenth of a cent (build), and continuous Doom at ~$7/hour (build). All numbers as reported by their authors. The pattern holds across dozens of independent builds: per-verdict cost rounds toward zero; only sustained high-frequency loops reach whole dollars.

The ecosystem's clients describe access running through the Vercel AI Gateway with the model id typesafe-ai/jev, which means the practical answer to "is it free?" is usually "is your gateway tier free, and for how many calls?" Check the official docs for today's truth, then do the napkin math from the cost table: items per day × questions per item × per-verdict cost. For most side projects the result is coffee money; for products, still coffee money, just daily.

One honest caution from the production guide: cheap is exactly when teams stop metering, and unmetered is how a runaway retry loop bills quietly for a month. Free-ish and monitored beats free and surprised.

Frequently asked questions

Is there a free tier for Jev?

TypeSafe's docs are the source of truth and launch-period tiers change; don't trust any third-party snapshot, including this one. What's stable is the reported workload economics above.

What does Jev cost per request?

Builder-reported figures put verdicts at fractions of a cent, with whole jobs (hundreds to tens of thousands of decisions) landing between $0.0004 and $0.13. Details: what builds actually cost.

Is Jev cheaper than GPT for my use case?

For decision-shaped work (classify, route, score, judge), reported numbers say dramatically yes; for generation, Jev isn't an option at all. The comparison draws the line.