Jev Pricing

/ 01

Short Answer

Start with two free live runs per day without signup, or five with Google sign-in. Paid VerdictKit plans add saved judges, history, CSV export and batch processing. These are browser workflow features, not API credits. The separate Jev model token-rate reference appears below.

/ 02

Our Plans

VerdictKit paid plans are purchased through the payment provider shown on your account page. Sign in to purchase on your account page; choose the plan before opening checkout. Access appears after payment is confirmed. The billing product name is VerdictKit, matching verdictkit.com. Free live runs remain available: two per day without signup or five per day with a Google account.

You pay for a browser workflow: save question configurations, rerun them, keep history and export results. Model token pricing is a separate cost reference below.

Subscriptions renew automatically until canceled. Review billing and refund terms before purchasing.

/ 03

From One Decision to a Reusable Workflow

1. Run your question in the playground and inspect the distribution.
2. With paid access, save the question configuration as a judge and reload it from your account.
3. On Batch, select the saved judge and upload a CSV with a state column.
4. Inspect the returned labels and probabilities, then download CSV results for review.

/ 04

Compare Plans

Trying a typed decision remains free. Paid access begins after the payment provider confirms payment.

/ 05

Storage and Operating Limits

Saved judges, batch records and paid run history remain until you delete them or request account data deletion; no fixed history expiry is currently configured. The account history screen displays the latest 100 runs. Up to 100 judges and 20 retained batch jobs are supported per account; delete an old item when a limit is reached.

Unlimited plans remove the run balance limit; request size, human verification, shared queues and upstream availability still apply. Individual runs allow 1–10 questions, at most 20 Choice options and a state under 10,000 characters. Batch supports 100 rows per file, a 10 KB per-row limit and four concurrent processing batches across the platform.

Free run content is not saved as history. See privacy and retention details.

/ 06

What Counts as a Run

This is the VerdictKit counting policy:

• One submitted evaluation counts as one run, whether it contains one question or up to ten.
• The two-run guest trial needs no signup. The five-run Free allowance is tied to a Google account, not an IP address; both reset each day at 00:00 UTC.
• Running a saved judge again counts as a new run.
• Each row in a batch file counts as one run. Batch has unlimited runs, so rows do not reduce a run balance.
• Requests that fail with a 4xx or 5xx response do not consume a run.

Demo previews do not consume runs. Live metering requires the configured server key, Turnstile and usage storage.

/ 07

The Jev Model Rate

Check the OpenRouter Jev 1.13 model page for the current public token rate. The listed rate matches TypeSafe’s direct API documentation at review time, but gateway platform fees can affect the cost of buying credits. For jev ai cost, distinguish model inference from gateway credit fees and VerdictKit plan prices. VerdictKit does not sell a customer-facing API gateway, resell API credits or provide a VerdictKit API key. Configured live playground requests pass through our server to OpenRouter; for your own API integration, use your own OpenRouter or TypeSafe account.

/ 08

Why Output Tokens Are Free

OpenRouter’s Jev 1.13 model listing charges for input tokens and lists output tokens at $0. Jev returns typed judgments — an option, a score or a yes/no probability — instead of generating prose. The API still reports output token usage, but the current model rate does not charge for those tokens.

/ 09

What One Call Actually Costs

The 476-token request is an illustrative size. Actual cost varies with state length and question count; the table multiplies that size by OpenRouter’s current $0.042-per-million model rate and excludes credit-purchase fees, infrastructure and other service costs.

/ 10

Jev vs an LLM on Price

TypeSafe’s use-case map describes a 100x cost advantage for a big-data scenario. A LangChain article also reports TypeSafe’s claim of up to 200x faster inference and 400x lower cost for classification tasks. Both figures are vendor claims about selected workloads, not independent universal benchmarks. Compare your own tasks, model choices and input sizes before projecting savings.

/ 11

How to Get More Out of Your Runs

1. Filter state first. Irrelevant detail can lower accuracy and adds input tokens.
2. Keep arithmetic in code; Jev 1.13 is weak at precise math.
3. If state cannot be filtered deterministically, a Noul can help screen relevance.

Read Jev’s known limitations before building a repeated workflow.

/ FAQ

Frequently asked questions