Jev FAQ: 20 Questions, Straight Answers

Jev is a "System One" AI model by TypeSafe that reads your input (state) plus typed questions and returns choices, scores, or true/false probabilities with confidence estimates — it never generates free-form text.

Choice (pick one option from a list you define), Score (an ordinal rating such as low/medium/high/critical), and Noul (TypeSafe's binary type: the probability that a statement is true, e.g. 0.95). Multiple questions can share one request.

A reference to Kahneman's Thinking, Fast and Slow: System 1 is fast, instinctive judgment. Jev handles the fast, bounded judgment calls (routing, scoring, approving) so that slow "System 2" reasoning — frontier LLMs or humans — is spent only where it matters.

No. Jev cannot write text, draft emails, summarize, or reason openly. It is a complement: it takes over the structured decision steps of an LLM stack so the expensive generative model is called less often.

Not necessarily. The official waitlist (typesafe.ai) has been clearing in waves — 140,000 users were cleared within about 36 hours of launch. Faster paths: the Vercel AI Gateway (typesafe-ai/jev), Cloudflare Workers AI (typesafe/jev), or OpenRouter (typesafe/jev-1.13) all provide keys with no waitlist.

Yes. Approved users get console.typesafe.ai/playground: paste text as state, write a plain-language question, pick the answer type, and get a sub-second result with a confidence score. You can mix Choice, Score, and Noul questions in one run.

Official Python and JavaScript/TypeScript SDKs ship from TypeSafe, plus framework integrations: the Vercel AI SDK (via @ai-sdk/typesafe) and LangChain (langchain-typesafe package).

Early-access defaults are reported at roughly 250,000 tokens per second and 1,200 requests per minute, with higher tiers available under enterprise agreements. Treat these as launch figures that may change.

The reference price is $0.042 per 1 million input tokens ($42 per billion), with output tokens unmetered and free — because the model returns structured decisions, not long text. A typical typed decision costs on the order of $0.0001 or less depending on state length.

Yes. Output is billed at $0 across all listed providers. The output is a compact typed decision (an option, a score, a probability), not a stream of generated tokens.

TypeSafe claims the $0.042/1M input rate is roughly 238x lower than "Claude Fable 5.1" and about 1/48th of GPT-5.6 Terra's $2.00/1M input — and those models also bill $2–$12 per 1M output tokens, which Jev doesn't charge at all. See our pricing page for a full comparison table.

Yes — and TypeSafe itself is candid about it: it cannot prove the price isn't subsidized and expects prices to fall rather than rise, but the current rate is a launch figure, not a guarantee. Our news page tracks price changes automatically.

It cannot generate text or explanations, cannot process images or audio (text only), cannot do arithmetic or date math (do that in code), and has a 64,000-token context window (32,000 on Cloudflare). It is also sensitive to irrelevant context — strip noise from the state before sending.

Precisely: it cannot produce text hallucinations because it never generates text, and its schema guarantee means it can only return one of the options you defined (0% type/structured-output errors by construction). But a valid answer can still be a wrong answer with high confidence attached — calibrate thresholds on your own data before automating.

A single HTTP call to the System One endpoint (e.g. POST https://api.typesafe.ai/v1/systemone with model jev-latest) carrying a state string and a map of typed questions; the response returns each answer with its probability. See our Get Access page for a copy-paste quickstart.

Reported end-to-end response times run 70–500ms, with community benchmarks around 200–260ms median per decision. TypeSafe's headline claim is 193.6x faster than text-generating models on System One workflows.

Treat them as promising but unproven. The 4-workflow benchmark (67.8% accuracy at $0.0004/case) was designed and run by TypeSafe, scored against consensus labels from GPT-6 Astra and Claude Fable 5.1 rather than ground truth — which TypeSafe itself admits biases toward OpenAI and Anthropic. No independent neutral-harness reproduction has surfaced yet.

An open-source community project that reproduces the Jev interface pattern on a small open model by reading typed option probabilities off the logits — evidence that the typed-decision primitive is durable even beyond one vendor.

TypeSafe AI is a startup that built Jev around a new training algorithm called Reinforcement Learning for Calibrated Decisions (RLCD). It entered the spotlight in September 2026 when 140,000+ people joined the waitlist within days of launch.

From TypeSafe's official site and docs, provider listings (OpenRouter, Cloudflare, Vercel), Hacker News and Reddit discussions, and independent write-ups — each news item links to its source. We mark all vendor-reported figures as self-reported and update this site automatically as new information appears.