Article
CommunityJev Guides & Articles
A Deep Dive into Jev
A practical tour with JavaScript examples: typed choices, scores, probabilities, real latencies and costs, the jagged edges, and how to split jobs correctly.
- Format
- Article
- Reading time
- 3 min
- Published by
- Community
- Author
- Flavio Copes
- Last verified
- Sep 2026
Works forJev for Scoring
Used inJev for Customer Support
Highlights
- Calls go to POST https://api.typesafe.ai/v1/systemone; 401 is a bad key, 422 invalid input, 429 rate limits, and 529 overload.
- Rate limits at writing: 250,000 tokens per second and 1,200 requests per minute; state plus all questions share about 64,000 tokens.
- A ~300-token support ticket costs about $0.0000126 per call at $0.042 per 1M input tokens, or $1.26 per 100,000 tickets.
- Speculative fan-out: 13 questions in one call were 12.2x cheaper and 10x faster than 13 sequential calls on a 53,777-character document.
- Adding one relevant example to a Score level moved a Safari bug from score 1.43 at 0.35 confidence to 1.03 at 0.96.
Quickstart
ts
import { choice, noul, score, TypeSafeClient } from '@typesafe-ai/sdk'
const client = new TypeSafeClient()
const { answers } = await client.systemOne({
state: { ticket },
questions: {
category: choice('What kind of ticket is `ticket`?', {
bug_report: 'Something is broken or behaving wrong',
other: null,
}),
has_repro_steps: noul('Does `ticket` say how to reproduce the problem?'),
},
})
console.log(answers.category.choice, answers.has_repro_steps.noul)Watch out
The 193.6x faster and 444.6x cheaper headline comes from TypeSafe's own workflow evaluations, which it calls the high end of real-world results, so treat it as a ceiling.
Questions about this entry
Short answers drawn from the linked project or page.
Mixed. The roughly 100 ms typical call and the $0.042 per million input token price are TypeSafe's own claims, and the loudest figure, 193.6x faster and 444.6x cheaper, comes from TypeSafe workflow evaluations the author himself calls the high end of real-world results, so read it as a ceiling.
The framing is API access, not a downloadable model. The post has dedicated sections on whether Jev is open source and whether it can be run locally, but the captured portion does not restate those answers, so read the article itself before planning around self-hosting.
About $0.0000126 per call for a roughly 300-token ticket at $0.042 per million input tokens, which works out to $1.26 per 100,000 tickets, and output tokens are free. Rate limits at the time of writing were 250,000 tokens per second and 1,200 requests per minute.
He keeps sections on where Jev breaks, on what people are building, and a set of skeptical notes about Jev and coding agents. One concrete result from his own use: adding one relevant example to a Score level moved a Safari bug from 1.43 at 0.35 confidence to 1.03 at 0.96, which is a prompt-design lever rather than a model fix.
Speculative fan-out is the big lever the post measures: 13 questions sent in one call against a 53,777-character document came out 12.2 times cheaper and 10 times faster than 13 sequential calls, with no change to any answer. The same point appears in the anatomy of a call, where a fourth question barely moves response time, so the marginal question is close to free once state is already sent.
Calls go to POST https://api.typesafe.ai/v1/systemone. A 401 means the key is bad, 422 means invalid input, 429 is a rate limit, and 529 is overload. The documented limits at the time of writing were 250,000 tokens per second and 1,200 requests per minute per account, and state plus all questions together share roughly 64,000 tokens, so a large state with many questions is bounded by that ceiling rather than by question count.
Route work before starting a coding agent, put a safety check in front of shell commands, prefilter the blog maintenance work, sort sponsor inquiries and newsletter replies, and add semantic priority to an events logger. He also asks whether Jev could cut an AI bill by 50 or 60 percent, which is the framing throughout: Jev is a component inside a regular application, not the interface or the worker.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Thread for Flavio Copes' practical Jev tutorial with JS examples.
2 pointsA deep dive into Jev, TypeSafe's System One modelRead the thread on Hacker NewsMore Jev resources like this
Tom Farnschläder's 24 September 2026 tutorial builds a support-ticket router with the Python SDK, using Choice, Score, and Noul in one call.
Jev Guides & ArticlesArticle#community#article#tutorial
Community
The Hacker News thread for Flavio Copes' JavaScript-oriented Jev tutorial covering choices, scores, latencies, and jagged edges.
Jev Guides & ArticlesPost#community#tutorial#javascript
Community
An evidence-backed community collection of Jev use cases, patterns, prompts, and starter code, with quick facts sourced to TypeSafe docs and runnable Python and TypeScript quickstart examples.
Jev Guides & Articles#community#awesome-list#tutorial
An interactive site by Sarah Drasner that explains System One thinking — fast, automatic judgments versus deliberate checking — and what that distinction means for software design, closing with the Jev × WebMCP Chrome extension as a case study.
Jev Guides & Articles#community#github#javascript
Community
sdras/system-one-explainerJavaScript
sdras/system-one-explainerJavaScriptCharlie Chen's five-minute guide to Jev: the hosted API opened to everyone on 21 September 2026, Choice, Score, and Noul in practice, a working Python request, and the jevmodel.org playground with free tokens.
Jev Guides & ArticlesArticle#community#article#tutorial
Community
A Chinese-language summary of the Latent Space interview with TypeSafe CEO Diogo Almeida, framing Jev as a production System One model rather than a general God model: parallel sampling, 70–500 ms latency, and $0.042 per million input tokens.
Jev Guides & ArticlesArticle#community#article#interview
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.