Jev Semantic Racer — race by typing words the AI judges
Each round shows a semantic condition. Type a word that fits, hit Race!, and Jev — TypeSafe AI's decision model — scores how well it fits with one real API call. The better the word, the further your car moves. Reach the flag before you run out of gas. No key, just play.
The same demo is one Jev call
- Show the target condition (e.g. opposite of HOT)
- The player types a word
- Send the condition + word to Jev as one
score - Jev rates the fit 0–3
- The car advances by the score
const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
method: "POST",
headers: { Authorization: `Bearer ${process.env.JEV_API_KEY}` },
body: JSON.stringify({
state: `Target: opposite of HOT. Word: "${word}".`, // the condition + the word
questions: {
fit: { type: "score", instructions: "How well does the word fit?",
criteria: ["miss", "weak", "good", "perfect"] },
},
}),
});
const { answers } = await res.json();
advance(answers.fit.score); // → 3 → "perfect"See a real decision perfect · 191ms
Target condition: a word that means the opposite of "hot". Player's word: "cold".
How Jev scores each word
Every word you submit is one real Jev API call. We send Jev the target condition and your word as a single typed `score` decision, with the criteria miss / weak / good / perfect. Jev returns a calibrated score from 0 to 3 — often a fractional value like 2.4 — in roughly 100–300ms, and your car advances by an amount mapped from that score. A perfect fit is a big jump; a near-miss barely nudges you forward.
Because each judgement is a normal API request, you see exactly what developers see when they build on Jev: the latency of the call and its cost (a fraction of a cent) are shown right under the verdict. It is the same score decision you would put inside a product to grade an answer, rank relevance, or gate a workflow — here it just happens to be judging your words.
What is Jev?
Jev is TypeSafe AI's first System One model. Instead of writing prose like a chatbot, it reads a situation and returns a typed, calibrated decision — a choice between options, a score on a scale, or a yes/no probability — in tens to hundreds of milliseconds, with no hallucinations. Judging how well a word fits a condition is exactly the kind of small, fast, bounded decision it is built for.
The same model powers content moderation, lead scoring, relevance ranking, ticket routing and agent tool-call guardrails in real products. Semantic Racer is just a fun, visible way to watch it decide.
Tips for a faster run
- Aim for the ideal answer, not just a related one — a perfect fit moves you far more than a weak one.
- Read the condition precisely: 'more intense than happy' wants ecstatic, not just glad.
- You have limited words before you run out of gas, so make each one count.
- Jev is calibrated, so an obviously spot-on word reliably scores high — trust clear answers over clever ones.
FAQ
Is Jev Semantic Racer free to play?
Yes. It runs in your browser with no API key and no signup — each word you submit is a free, rate-limited call to the live Jev decision API.
Does Jev really judge every word?
Yes. Each submitted word is sent to the Jev API as a single typed `score` decision against the round's condition, and Jev returns a calibrated 0–3 score. Nothing is scripted or pre-computed.
How does my car move?
Your car advances by an amount mapped from Jev's score: a miss barely moves you, a perfect fit is a big jump. Reach the finish line before you run out of words to win.
How fast and how expensive is each judgement?
Each word is a single API call that typically returns in about 100–300ms and costs a fraction of a cent in input tokens — the page shows the live latency and running cost.
More: Moral Dilemmas — you vs Jev · Jev Gomoku · All decision games · Jev API docs