Independent Jev coverage
Explore the experiments, access options and open questions behind Jev.
Jev is TypeSafe AI’s decision model: it evaluates context and returns structured choices, scores and probabilities instead of writing a chat reply.
Think of choosing where a request goes next, rather than composing the answer.
View sourceA headline result needs a task, a baseline and a source.
193.6×
TypeSafe’s workflow evaluation headline, not a result for every task.
The provider describes these gains as being at the high end of expected real-world gains. Not our measurement.
Match the task, input length, location, baseline and reasoning settings before comparing results.
Compare accuracy and completed-task cost alongside speed. Results can vary by workload.
How we handle evidenceUnderstand the decision, the surrounding code and the limits.
Moritz Kremb built a voice-controlled browser. Jev selects the intent and target from partial transcripts, while Playwright performs the browser action.
The repository describes the implementation and tests, but Jev Tracker has not run the browser or repeated its results.
Element snapshots are capped and do not include content inside iframes; destructive actions still depend on a spoken confirmation policy.
Compare provider routes, account requirements and published prices.
| Channel | Status & requirements | Published price | Source checked | Next step |
|---|---|---|---|---|
| TypeSafe AI Official jev-latest | Waitlist The public site still offers a waitlist. Documentation also links to a sign-in-gated Playground; admission was not tested. A TypeSafe login is required for the Playground. Account eligibility and waiting time are not yet verified. | $0.042 per 1M input tokensView source | Stale: more than 72 hours since the source check. Confirm with the provider. | Open PlaygroundView source |
| Vercel AI Gateway Integration typesafe-ai/jev | Announced available Vercel announces Jev support and lists the model. Team eligibility and successful calls were not tested. Gateway credentials and an integration capable of evaluation calls. Confirm access and billing with Vercel. | $0.04 per 1M input tokensView source | Stale: more than 72 hours since the source check. Confirm with the provider. | Visit AI GatewayView source |
Separate the product’s positioning from evidence of what it can reliably do.
TypeSafe calls its approach “System One.” That is the provider’s positioning, not a settled independent classification.
Our editorial question: which capabilities, evaluation methods and task outcomes distinguish it in practice?
View sourceThe provider documents that Jev is not trained to generate text. Chaining choices into prose is not a practical substitute.
For jev-1.13.0, the model reference specifies text-only input: strings, JSON objects or arrays of text values. Images, audio and video are not accepted directly.
TypeSafe derives confidence from the output probability distribution. Its guidance makes action thresholds depend on the domain, risk and observed performance.
Announcements and integration updates.
An integration walkthrough places Jev inside an agent loop.
The integration exposes typed evaluations rather than chat completions.
The launch post presents System One models and early demonstrations.
Original sources, review dates and context for each claim.
We inspect the cited material and distinguish its claims from our interpretation. No experiment on this page was independently run by Jev Tracker.
We use Google Analytics to understand site usage. Read our privacy notice for details about analytics, browser preferences and email contact.
Read our full editorial methodWhen we correct a material error, we explain what changed, why and when. Send a report by email using the guidance on our Corrections page.
Corrections and updates