Can I host it myself or get the weights?

No. The post lists downloadable weights and a license as not yet confirmed, with access limited to TypeSafe's early-access API. It also leaves the context window unconfirmed, so do not plan capacity around a number you cannot verify.

Are the 193.6x faster and 444.6x cheaper figures measured?

They are TypeSafe's published comparisons on selected workflows, not Kie.ai measurements, and the post says so. It also notes TypeSafe has not published enough technical detail to reconstruct the architecture, parameter count or hardware.

Does bounded output mean I cannot get a malformed answer?

You cannot get an invalid option or broken JSON for that field, since the answer space is declared up front. But it can still pick the wrong valid option or attach an overconfident probability, so the schema guarantee is not an accuracy guarantee.

Does it explain its reasoning or write the reply for me?

No to both. System One models do not write replies, produce code or generate explanations, so the model supplies the judgment while your application owns thresholds, policy, retries and any customer-facing text.

Is a context window or parameter count published?

No. The article's at-a-glance table lists the context window as not yet confirmed, and says TypeSafe does not publish enough technical detail to reconstruct the full architecture, parameter count or hardware implementation.

What does the published pricing work out to?

$0.042 per 1 million input tokens with output free, which equals $42 per 1 billion input tokens. That figure excludes application-side costs such as storage, orchestration and data preparation.

How was Jev trained, according to the article?

After two years in stealth under founder Diogo Almeida, with an approach TypeSafe calls Reinforcement Learning for Calibrated Decisions, or RLCD, intended to make confidence values useful as operating signals.

CommunityJev Guides & Articles

What is Jev?

Kie.ai on Jev as a System One evaluation model: parallel typed questions, pricing, and where TypeSafe says hosted weights are not confirmed yet.

See highlights

Open kie.ai

Format: Article
Published by: Community
Last verified: Sep 2026

Tagscommunity article explainer pricing

Works forJev for Classification

Used inJev for Software & DevTools

Highlights

Release dated 15 September 2026 at $0.042 per million input tokens, output free.
TypeSafe describes 70 to 500 millisecond response times for System One tasks.
Restates Choice, Score, and Noul and parallel evaluation on one state.
Notes early-access API access while downloadable weights are not yet confirmed.
Repeats TypeSafe workflow comparisons up to 193.6× faster and 444.6× cheaper on selected tasks.

Watch out

The speed and cost multiples are TypeSafe's published comparisons, not Kie.ai measurements.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

X: New signups no longer get free credits.

Hacker News: Simon Willison's overview of Noul, Choice, and Score, and where a decision model fits.

benwerd53 pointsJev introduces a new shape of LLMRead the thread on Hacker News

Hacker News: swyx posts the Latent Space interview with Diogo Almeida.

swyx4 pointsJev: System One Models for Prod, Not God – With Diogo Almeida, CEO, TypeSafe AIRead the thread on Hacker News

X: Diogo Almeida in five minutes

Related terms

Glossary definitions related to this entry.

Token pricing

More Jev resources like this

TypeSafe Jev explained

Requesty's September 2026 guide to Jev as a System One model: state, Choice, Score, and Noul, plus how Requesty routes typesafe/jev-1.13.0.