What is Jev?
Jev is the first System One decision model, released by TypeSafe AI on September 15, 2026. You send it text plus typed questions; it returns typed answers with probabilities and a confidence score. It does not generate prose, code, or explanations — it only decides.
Updated September 2026 · 5 min read
In short
- A frontier model optimized for decisions, not text generation
- 20–200x faster and 40–400x cheaper for the same judgment calls (output tokens are free)
- Answers are typed: an option, a score, or a true/false probability — never a paragraph to parse
- Per-question cost is so low that agents can ask 60 questions for the price of one LLM call
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
The launch announcement from Diogo Almeida, Jev's creator.
How a call works
One endpoint, one request: a state (the thing being judged) plus a map of questions. Three question types, mixable in a single call and evaluated in parallel:
| Type | Returns | Use for |
|---|---|---|
| choice | One of up to 255 labelled options, plus per-option probabilities | Routing, classification, picking |
| score | A fractional value on an ordered 2–10 spectrum, plus the distribution | Severity, quality, sentiment |
| noul | A calibrated 0–1 yes/no probability | Gates, filters, guardrails |
This site itself uses Jev for submission review: paste a link, Jev decides whether it belongs (noul) and picks the use case (choice) — one call, fractions of a cent.
Speed, cost and accuracy
| Model | Accuracy | Cost / case | Time / case |
|---|---|---|---|
| Claude Opus 5 | 73.1% | $0.1761 | 37.8 s |
| Jev | 67.8% | $0.0004 | 0.07–0.5 s |
| Terra | 67.9% | $0.0304 | 10.1 s |
| Claude Haiku 4.5 | 53.6% | — | — |
Jev matches Claude Sonnet 5-class judgment accuracy at ~1/290th the cost. Caveat, honestly stated: Reference answers in these evals came from two large models, not humans. The widely-quoted input price is $0.042 per million tokens (the launch announcement); TypeSafe's own site lists $0.25–$0.42/M. Output tokens are free either way. We flag the discrepancy here →
Models and limits
- Models: jev-latest, jev-preview
- Context: 64k tokens (32k is quoted in some vendor docs)
- Rate limits: 250k tokens/s, 1,200 requests/min
- Response: typically 70–500 ms
Where it does not fit
- It writes no text — summaries, code, and replies are out of scope
- Options must be predefined; it cannot invent choices
- Text-only input (no images or audio)
- Questions are independent — no multi-step dependent reasoning inside one call
Why it is called Jev
From the Jevons paradox: when a resource gets radically cheaper, consumption explodes instead of shrinking. Make decisions 400x cheaper and agents stop rationing judgment — they ask everything. “System One” nods to Kahneman's fast, automatic mode of thinking: not deliberation, instant judgment.
What people build with it
Browser agents that search and book flights for $0.004 a run. Game agents holding 60fps decision loops. Inbox triage that reads 400 emails a day. Trading backtests where Jev plays the discretionary trader. Catalogs of 1,000+ papers classified for $0.08. The community ships new builds daily — the best of them live in our directory, ranked by upvotes.
Common questions
Is Jev an LLM?
Can Jev hallucinate?
How is it so cheap?
How do I try it?
/v1/systemone. Or submit a link on this site — our reviewer is Jev itself.JevPicks is an independent community project, not affiliated with TypeSafe AI. Specs and eval numbers come from TypeSafe's published materials; directory figures come from this site's own submissions and are updated as builds arrive.