Jev cost planning
Jev cost planning is how you turn the public list price into a month you can defend: measured input tokens × calls × vendor unit price × (1 + gateway uplift). Snapshot date: 2026-09-21. Jev is TypeSafe’s System One decision model. This page does not invent traffic.
Use it after you have read the canonical sticker on Jev pricing. Compare frames: pricing explained. Free-tier honesty: Is Jev free?.
Formula (yours to fill)
monthly_$ ≈ (avg_input_tokens_per_call / 1e6) × calls × unit_price × (1 + gateway_uplift)
avg_input_tokens ≈ filtered state + all question text in that request. Official: state is ingested once per request — batch questions on the same state.
Do not invent calls. Use last month’s ticket count or a measured pilot. Log usage.input_tokens and response.model.
First-party vendor unit price on the models page: $0.042 per million input tokens, output free.
Channel unit prices (vendor cards, 2026-09-21)
| Door | Unit price we saw | Caveat |
|---|---|---|
| TypeSafe first-party | $0.042 / M input; output $0 | Waitlist; no public self-serve free tier on the models page |
| Vercel AI Gateway | Card repeats $0.042 / 1M input | Gateway/platform fees extra |
| OpenRouter | Card repeats $0.042 / $0 | Needs OR credit; 402 is theirs |
| Cloudflare Workers AI | “View pricing in dashboard” | Context card 32k |
Higher limits: TypeSafe says contact sales. We will not invent enterprise SKUs.
Worked examples (illustrative — replace with yours)
Pilot on first-party HTTP. 800 input tokens/call × 200,000 calls/month:
800 / 1e6 × 200_000 × 0.042 ≈ $6.72 / month in TypeSafe list token fees.
One triage call. 450 input tokens:
450 / 1e6 × 0.042 ≈ $0.000019 per call.
Vendor demo, not a forecast. TypeSafe’s launch post says an engineer worried about 10 queries/second on a Doom bot and that it “ends up costing ~$7/hour.” That is their payload and rate, dated 15 September 2026.
Add retries, playground clicks, failed 422s you still paid input for, and gateway markup. Arithmetic on assumed volume is not a promise.
Levers that are documented
- Filter state (money and 1.13 accuracy).
- Many questions, one state. TypeSafe’s parallel-questions cookbook publishes a vendor-measured 12.2× cheaper example vs N calls on their payload — do not stamp that multiple on your traffic.
- Pin versions so an alias move does not change token use silently.
- Do not generate with Jev (it cannot; chaining Choices is the expensive failure mode they warn about).
- Watch RPM vs token/second separately — rate limits. Cheap tokens still 429.
What a finance partner should see
A dated sheet: door, unit price source URL, measured tokens/call, measured calls, retries, gateway uplift, and a note that output is free per TypeSafe’s card. No invented “industry average volume.” If they want 193× / 444× savings, send them the vendor blog and your harness, separately. Jev review explains why those multiples are not a price list.
FAQ
Are output tokens billed?
TypeSafe models page (2026-09-21): output free; usage.output_tokens still appears.
Is there a free tier I can put in the model?
Not described on the public models page we used. Third-party promos expire. Is Jev free?.
Does jev.pro sell tokens or committed-use discounts?
No.
Should I budget the LLM invoice or the Jev invoice?
Budget the path: Jev for the snap decision, an LLM if you still draft. Pricing explained separates the two meters.
What this page does not claim
- No search-volume or “customers save 80%” claims.
- Launch-post commentary that prices may be subsidized is theirs.
- Open replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed.