Access · Last updated 2026-09-21

Jev pricing

Jev pricing is the vendor list price TypeSafe publishes for Jev (TypeSafe’s System One decision model). Snapshot date: 2026-09-21. The models page is the source of truth: if a blog, gateway card, or this page disagrees, docs win.

This hub is for builders who need a dated sticker price, a cost-per-call sketch, and an honest list of what TypeSafe has not published. Satellites: Jev channels, Is Jev free?, Jev pricing explained, where to buy, pricing by channel, and the cost planner.

List price (vendor claim, 2026-09-21)

TypeSafe’s models table for Jev 1.13 (jev-1.13.0) lists:

$42 per billion input tokens ($0.042 per million / MTok)
Output tokens are free. Docs: “Charged per input token. Output tokens are free.”
Only input tokens are billed. The API still reports usage.output_tokens for debugging.

The 15 September 2026 launch post states the same unit price and calls output “FREE (too cheap to meter).” It also says TypeSafe cannot prove the price is not subsidized and expects prices to go down, not up. That is their commentary, not a contract.

Gateway cards we checked the same day repeat the unit price unless a promo row says otherwise; they do not replace the models page:

OpenRouter Jev 1.13: $0.042 / $0 per 1M, plus a weighted avg input ~$0.04152 / M that they publish.
Vercel model card: $0.042 / 1M input. Same-day provider table: Free/Free. Rival notes cited a promo through 25 September 2026 — verify the dashboard.

Gateway or marketplace platform fees, credits, and invoices are those vendors’ bills. Full matrix: pricing by channel.

Worked cost examples (illustrative volumes)

Formula (first-party list price, no gateway markup):

cost_$ ≈ (input_tokens / 1_000_000) × 0.042 × calls

Replace input_tokens and calls with your measured usage. Do not treat the volumes below as traffic promises.

Sketch	Assumed input	Assumed volume	List-price arithmetic
One typical triage call (one short message + a few questions)	450 tokens	1 call	`450 / 1e6 × 0.042 ≈ $0.000019`
Pilot month	800 tokens / call	200,000 calls	`800 / 1e6 × 200_000 × 0.042 ≈ $6.72`
Launch-post Doom demo	vendor’s “10 queries a second”	~1 hour	TypeSafe wrote “~$7/hour” for that demo — their example, not ours

State is ingested once per request. Batching many questions on the same state is the documented way to cut tokens versus N separate HTTP calls. TypeSafe’s parallel-questions cookbook publishes a vendor-measured 12.2× cheaper example on their payload — do not copy that multiple onto your traffic.

Add retries, playground traffic, and any gateway uplift yourselves. Cost planning walks the sheet a finance partner should see.

Cheapest path today (2026-09-21)

Sticker and effective cost diverge. Dated tree — re-run it when a card moves:

Vercel dashboard still shows Free for typesafe-ai/jev (provider table did today; promo notes cite 25 September) and you already have a Vercel team + card → setup: Jev on Vercel AI Gateway. Do not buy credits if you still want the monthly free-credit bucket. Watch free-tier 429s.
No TypeSafe invite, need unified billing → setup: Jev on OpenRouter. List $0.042; OR publishes ~$0.04152 weighted avg input. Same TypeSafe upstream on their provider table.
Console key in hand, want official pin + 64k envelope → setup: Jev on TypeSafe.
Worker-native → Cloudflare (dashboard $/M only).
Netlify credits → models are listed at $0.04 / $0.00; confirm System One actually routes.

Setup hub: Jev channels. Storefronts and “not found on Together / Fireworks / Groq”: where to buy.

Gateway vs first-party

Same unit sticker does not mean the same invoice.

TypeSafe HTTP (POST /v1/systemone): billed at the models-page list price after you have a console key. Early access / waitlist — see get access.
Vercel AI Gateway / OpenRouter / other doors: cards often echo $0.042 / M input unless a promo row is $0. You still pay that door’s account, budgets, and any platform markup. A 402 on OpenRouter is their credit error.
Cloudflare Workers AI: the public card we used says “view pricing in the dashboard,” and lists a 32k context — narrower than TypeSafe’s 64k request budget.

Do not mix a TypeSafe bearer token with a gateway base URL and file the 401 as a pricing bug.

What is not documented (do not invent)

The models page snapshot we used does not describe:

a self-serve free tier
trial credits or a “starter” quota
committed-use or volume discounts
per-seat SaaS plans named “Pro” (on this site, Pro means production craft — see about jev.pro)

Higher rate limits are “available on custom and enterprise plans” via [email protected]. We will not invent enterprise SKUs or unpublished list prices.

Rate limits published the same day (also vendor, and TypeSafe says they can change without notice): 250,000 tokens/second and 1,200 requests/minute. Details: rate limits.

Cost levers that are documented

Filter state. Unused context costs money and, on 1.13, accuracy (jaggedness).
Many questions, one state. Do not fan out one-question HTTP calls.
Pin jev-1.13.0 (or the version you fitted) so an alias move does not change token use silently — models and aliases.
Do not try to generate text with Jev. It cannot; chaining Choices as a decoder is the expensive failure mode TypeSafe warns about.

FAQ

What is Jev pricing right now?

According to TypeSafe docs on 2026-09-21: $0.042 per million input tokens, output free, for Jev 1.13. Re-read docs.typesafe.ai/models before you bid a contract.

Is Jev free?

Not as a documented self-serve free product. Output tokens are free; input tokens are billed at the list price. Early access and third-party gateway promos are not a TypeSafe free tier. Full answer: Is Jev free?.

How much does one Jev call cost?

(your input tokens / 1e6) × 0.042 at first-party list price, plus any gateway markup. A 450-token triage sketch is about two hundredths of a millidollar. Measure usage.input_tokens on your payloads.

Are “400× cheaper” homepage figures included in the list price?

No. TypeSafe’s workflow-eval multiples (they cite 444.6× cheaper as a high-end example) are vendor-reported comparisons, not a second price list. Label them. Jev vs LLM billing separates sticker price from those claims.