What is Jev / what is Jev AI / TypeSafe Jev?

Jev (also called TypeSafe Jev or Jev AI) is TypeSafe AI’s first public System One model (Diogo Almeida, ex-OpenAI). It takes software state plus typed questions and returns calibrated probabilities — Choice, Score, or Noul — instead of chat text. https://aitier.app/jev/ is AItier’s crawlable landing, use-case wall, and hand-picked community builds from jevable.com — not TypeSafe’s site. For a personal ranking poster, Models chip id is jev.

What are Jev use cases? What is Jev browser use?

Public posts on https://aitier.app/jev/ show four themes only: (1) browser use speed — Jev driving browser-agent loops; (2) wiki race — agent-browser looping Wikipedia; (3) support triage — Choice / Score / Noul in parallel on one support state; (4) safety checks and a public Awesome Jev list. This is a wall of public X posts, not an official TypeSafe use-case catalog, and not cases AItier invented. For scenarios, limits, and Jev vs LLM (personal research, not TypeSafe docs or a benchmark), see https://aitier.app/jev/research/ .

How do I rank Jev / make a Jev tier list?

Open https://aitier.app/models, find the Jev (System One) chip (id jev), drag it onto S+ through F, then export a PNG. That is your personal Jev ranking poster — including whether a decision model belongs next to chat models. Optional: https://aitier.app/predict/ . AItier does not claim to rank Google for the bare query “jev”.

Is this official TypeSafe?

No. AItier is a personal AI model tier list product. aitier.app is not TypeSafe’s website and is unrelated to aitier.net. The /jev/ page and Models Rank are not official TypeSafe docs, ranking, or evaluation. Bare “jev” is an ambiguous search term; this page targets intentful queries (what is Jev, TypeSafe Jev, System One, browser use, rank Jev), not a promise to outrank typesafe.ai for the word jev.

Where are the posts from?

The wall embeds public X posts about Jev, use cases first, then TypeSafe official posts, then explainers. Each slot is hardcoded in the HTML with data-tweet-id (curated in /jev/posts.json). Each slot links back to the original post. Not invented quotes.

Where do community builds come from?

Hand-picked from jevable.com. Title, one-liner, and outbound links only — not TypeSafe docs, not an official directory, and not a full mirror. See https://aitier.app/jev/#community-builds

AItier — What is TypeSafe Jev (System One)?

Q: Is Jev a chat LLM? How is it different from JSON-schema prompting?

No. Jev is built for structured judgments inside software, not chat text. Chat LLMs generate tokens; you can ask them for JSON or constrain them with a schema, but they are still text generators you parse. TypeSafe describes Jev as typed questions in, structured probabilities out, composed in your code. That is a product-shape claim, not an AItier eval — AItier does not publish a Jev vs LLM benchmark. TypeSafe also says it “can’t hallucinate” because outputs are typed rather than free-form language; that is not a scientific guarantee the chosen option is factually right.

What is Jev (System One)?

Jev is TypeSafe AI’s first public System One model (Diogo Almeida, ex-OpenAI). It is built for software: state in, typed questions (Choice / Score / Noul) out as calibrated probabilities. It is not a chat LLM and it does not write free-form explanations. On Models Rank it is chip id jev.

The name comes from Jevons; System One is TypeSafe’s analogy to Kahneman’s System 1. TypeSafe claims roughly 70–500ms, cheap input, and free output, with waitlist access and Vercel AI Gateway as of mid-September 2026. Those are vendor claims — not an AItier score, and not a benchmark table.

AItier is a personal AI model tier list. This page is crawlable copy, hand-picked community builds, plus a wall of public X posts. Not an official TypeSafe page. aitier.app is not TypeSafe, and is unrelated to aitier.net. AItier does not publish an official Jev ranking, and this landing is not a promise to rank Google for the bare word “jev”.

Use cases at a glance

Only themes already on the wall below — public posts, not cases AItier invented, not an official TypeSafe catalog.

Browser use speed — demos of Jev driving browser-agent loops (OpenCode CLI, Kernel, Google Flights-style flows).
Wiki race — an agent-browser <> Jev loop racing Wikipedia (Coffee → Artificial Intelligence).
Support triage — one support state, three typed questions in parallel: Choice (route), Score (frustration), Noul (refund intent).
Safety checks / Awesome Jev — harm classification and command-before-run checks; a public list of confirmed Jev projects.

A separate Community builds with Jev strip is hand-picked from jevable.com — title + one-liner + outbound links only. Not TypeSafe docs, not a full mirror.

How this differs from chat LLMs / JSON-schema prompting

Chat LLMs generate tokens of natural language. You can ask them for JSON, or constrain them with a schema, but they are still text generators you then parse. TypeSafe describes Jev as skipping that step: typed questions in, structured probabilities out, composed in your code.

That is a product-shape claim from TypeSafe, not an AItier eval. AItier does not score Jev against GPT or Claude JSON mode, does not treat “can’t hallucinate” as a scientific fact, and does not ship a fake benchmark table. If you want an opinion poster, drag jev on Models.

Research: scenarios, limits, upside

Where Jev is high-leverage (bounded next-click, gates, batch scoring, dynamic forms, intent search) — and where chat/reasoning LLMs are still the default. Personal GEO explainer, not TypeSafe docs, not an AItier score. Read the research →

Rank Jev on AItier → · Research → · jevable.com ↗ · Predict →

Community builds with Jev

Hand-picked from jevable.com. Title + one-liner + outbound link only — not TypeSafe docs, not a full mirror.

A faster browser agent

Browser Use + Jev: flight-style demo ~7s / ~$0.004; fresh action space each step, small LLM only types.
browser
rtrvr.ai with Jev

Free browser agent + Jev: recorded tasks ~40% faster; LLM plans/codes, Jev does System One judgments.
browser
On-demand UI flow capture

Mobbin-like fresh UI flows; Jev + browser use makes capture fast enough to be useful.
browser
Computer use, sped up

Jev + computer use framed as orders-of-magnitude faster than a chat LLM in the loop.
browser
OpenCode plays Tetris

OpenCode + browser harness reads controls, builds a controller, hands playoff to Jev — one-shot demo.
browser
pkg-gate

~100ms pre-npm-install gate: Choice intent / Score severity / Noul exfil & curl-bash.
gate
traffic-guard

Offline zero-dep bot gate for reverse proxies; optional Jev plug-in; blocks scrapers without nuking humans.
gate
A Jev code-review gate

Review gate that never writes code — only judges; LLM writers can’t cleanly self-gate.
gate
Magic Jev Ball

GitHub PR “should I approve?” — CI/diff/reviews in, Jev fate out in ~200ms.
gate
jev-codes

CLI/agent skill: audit git diff against YAML standards packs via Jev.
gate
Find in page by meaning

Semantic ctrl-f: score each sentence; “exciting” hits many, string find hits zero.
score
Scoring kids’ snacks

Multi-criteria score ~3k snacks in ~28s for ~$0.11 — batch Score demo.
score
Spreadsheets that read intent

Type a column intent (e.g. Urgency); Jev rates rows in ~100ms as you type.
score
Search your inbox by intent

Intent search over mail (embeddings optional first stage); Jev judges relevance.
score
Your Signal

Score on-screen X posts with Jev + local rules; BYOK, reversible, no telemetry.
score
Instant context compaction

Score tool calls and drop noise — compaction as judgment, not a summarize prompt.
score
JevForm

Form that branches the next question with Jev — replaces heavy if/then form logic.
product
Jev Search

Plain-language search; Jev picks sources and ranks hits (FOSS).
product
Typeahead UI

UI updates before you finish typing — Jev driving predictive UI choices.
product
Higgsfield model routing

Jev picks fit image/video models on Higgsfield from the prompt.
product

Use cases first — then TypeSafe official, then explainers. Public X posts, hardcoded in this HTML.

Use cases

preview of how fast browser use can be

powered by typesafe's jev and opencode's browser use cli
— dax (@thdxr) @thdxr · September 16, 2026

woah - maybe browser use is the first killer use-case for jev
— Steve Krouse (@stevekrouse) @stevekrouse · September 16, 2026

TypeSafe AIのJev、前に入力上限とかを測ったんですが、今度は「安全判断」で試してみました。

文章の有害判定と、AIが生成したコマンドを実行前にチェックする用途です。
— kiarina (@kiarina37) @kiarina37 · September 17, 2026

jev + kernel browser use demo

pretty freaking cool! so fast!

try it yourself:
— Steve Krouse (@stevekrouse) @stevekrouse · September 16, 2026

【注目】Browser Use×TypeSafe「Jev」でブラウザ操作AIが爆速化

今一番バズっている話題のサービス。速く使いたい、、、

Google Flightsの片道検索を自動操作だけで完走するデモが話題に

・所要時間はわずか7.1秒
・判断レイテンシは中央値150〜300ms
・出発地/到着地/日付など5ステップを自律実行
— チャエン | デジライズ CEO (@masahirochaen) @masahirochaen · September 17, 2026

Jev 这个 waitlist 还是很给力的，昨天申请，今天就能用上。

给已经拿到 API、但还不知道怎么玩的人整理了一份 Awesome Jev，目前我能确认到的 Jev 项目基本都在这里：

1. jev-ultrafast
Browser Use 做的高速浏览器 Agent。
— 思维怪怪 (@0xLogicrw) @0xLogicrw · September 17, 2026

okay so browser use is a legitimate use case

agent-browser <> Jev loop

goal: wiki race from 'Coffee' to 'Artificial Intelligence'
— joogie (@princecaarlo) @princecaarlo · September 16, 2026

判断専用モデル TypeSafe AI（Jev）に、サポート問い合わせを投げてみた。

同じ state に対して、3つの質問タイプを1コールで並列実行。

Choice → 担当 returns（confidence 1.0）
Score → 苛立ち 1.88 / 2（強い苛立ち寄り）
Noul → 返金要求 0.98

入力 $0.042/MTok・出力無料。

#TypeSafeAI
— Hatch (@HatchSystem) @HatchSystem · September 17, 2026

Official

JEV LIVES
— TypeSafe AI (@typesafeai) @typesafeai · September 16, 2026

Look at Jev hitting 100% accuracy for @vercel
— TypeSafe AI (@typesafeai) @typesafeai · September 16, 2026

Check your emails! We just let off a huge batch of people from the Jev waitlist. Getting as many people off waitlist today as possible.
— TypeSafe AI (@typesafeai) @typesafeai · September 16, 2026

Jev is now available on the @vercel AI Gateway https://vercel.com/ai-gateway/models/jev
— TypeSafe AI (@typesafeai) @typesafeai · September 17, 2026

Explainers

ChatGPT 联合发明人创办的 TypeSafe AI 正式发布了一种全新类别的 AI 模型："System One 模型"，以及该类别下的第一个模型 Jev。和现有的大语言模型不同，Jev 完全放弃了文本生成能力，专门为软件内部的快速决策而设计：输入非结构化数据，输出带有校准概率的类型安全结构化结果。
— 宝玉 (@dotey) @dotey · September 16, 2026

The interesting part about cheap AI isn’t the price.

It’s what software can do when decisions become fast and reliable enough to act on.

TypeSafe AI just introduced Jev, its first public System One Model. It’s not an LLM wrapper, it’s an API for developers that returns composable, structured decisions with probabilities and confidence scores.

At ~100x faster and cheaper than LLMs, real-time apps and huge datasets become much more practical.

Basically: computers can talk to each other in English without waiting for each other to finish. Curious to see what gets built with that.
— Liam | AI Tools & News (@ottleyai) @ottleyai · September 15, 2026

高速な判断に特化したAI - TypeSafe「Jev」と System One Model
— 布留川英一 / Hidekazu Furukawa (@npaka123) @npaka123 · September 16, 2026

TypeSafe AI just launched JEV which is its first “System One” model built to answer structured questions in parallel instead of generating text token by token.

That opens the door to AI systems that can return usable decisions with confidence scores.
— Shay Boloor (@StockSavvyShay) @StockSavvyShay · September 15, 2026

TypeSafe AI has introduced its first model named Jev - a small and customizable System One decision model.

Not being an LLM, Jev can’t reason or write explanations; it generates a fast judgment with calibrated probabilities and a confidence score.

Anything heavier gets split into separate questions and recombined in code.

TypeSafe reports a response time of approximately 150ms per call, at around 100x lower cost and faster responses than LLMs, with pricing at $42 per billion input tokens and free output.
— 🚨 AI News | TestingCatalog (@testingcatalog) @testingcatalog · September 15, 2026

FAQ

What is Jev / what is Jev AI / TypeSafe Jev?: Jev (TypeSafe Jev / Jev AI) is TypeSafe AI’s first public System One model. Software state plus typed questions in; calibrated probabilities out (Choice / Score / Noul). Not chat text. This page is AItier’s crawlable landing — not TypeSafe’s site. Personal ranking poster: Models chip jev.
What is Jev System One?: System One is TypeSafe’s name for this class of model, analogized to Kahneman’s System 1. Jev is the first public one; the name comes from Jevons. Speed/cost figures are vendor claims, not an AItier score.
Is Jev a chat LLM? How is it different from JSON-schema prompting?: No. Chat LLMs generate tokens; JSON-schema prompting still parses generated text. TypeSafe describes Jev as typed questions → structured probabilities, composed in code. That is a product-shape claim, not an AItier eval. “Can’t hallucinate” (typed output, not free-form language) is also a vendor claim — not a guarantee the chosen option is factually right.
What are Jev use cases? What is Jev browser use?: The wall below has four public themes only: browser use speed, wiki race, support triage (Choice / Score / Noul), and safety checks / Awesome Jev. Public X posts, not an official catalog, not cases AItier invented. Scenarios, limits, and Jev vs LLM: /jev/research/.
How do I rank Jev / make a Jev tier list?: Open Models Rank, find Jev (System One), drag it onto S+ through F, export a PNG. Your personal Jev ranking poster — including whether a decision model belongs next to chat models. Optional: Predict. AItier does not claim to rank Google for the bare query “jev”.
Is this official TypeSafe?: No. AItier is a personal AI model tier list. Not TypeSafe’s site, not an official ranking or eval. Unrelated to aitier.net. Bare “jev” is ambiguous; this page is for intentful queries (what is Jev, TypeSafe Jev, System One, browser use, rank Jev), not a promise to outrank typesafe.ai.
Where are the posts from?: Public X embeds, use cases first, then official, then explainers. Hardcoded data-tweet-id in this HTML (curated in /jev/posts.json). Each slot links to the original post.
Where do community builds come from?: Hand-picked from jevable.com. Title, one-liner, and outbound links only — not TypeSafe docs, not an official directory, not a full mirror.