Jev API

Jev API — no waitlist, instant access

TypeSafe's System One model. Send program state + typed questions, get back structured decisions in 70–500ms. Not text generation — typed, probabilistic answers you can branch on in code.

Latency 70–500ms Price $0.042 / M tok 0% structured-output errors
Jev API Key Quick Start ↓ Three Primitives ↓

What is the Jev API?

Jev is a frontier System One model from TypeSafe AI, announced September 15, 2026. Built by Diogo Almeida (co-inventor of RLHF and InstructGPT at OpenAI), backed by $40M led by DCVC.

Unlike LLMs that generate text token-by-token, Jev evaluates your program state against a map of typed questions and returns structured answers — one per question — in a single parallel pass. That means:

JevFrontier LLMs
End-to-end latency 70ms – 500ms 3s – 329s
Input price $0.042 / M tok $0.20 – $10 / M tok
Output price free ~5× input
Structured-output errors 0% (by construction) 0.58% – 45.5%

Source: TypeSafe AI & independent benchmarks. Jev is available in early access — you need a waitlisted API key. You can also access it through OpenRouter, Vercel AI Gateway, Netlify AI Gateway, and AIMLAPI.

The endpoint

One endpoint. Three question types. That's the entire API surface.

POSThttps://tokenra.io/v1/decisions
Authorization: Bearer <API_KEY>
Content-Type: application/json
Python — Quick Start
# Install: pip install requests
import requests

url = "https://tokenra.io/v1/decisions"
headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json"
}

payload = {
    "model": "jev-latest",
    "state": "Help! My payouts have been failing for 3 days.",
    "questions": {
        "is_urgent": {
            "type": "noul",
            "instructions": "Does this message convey urgency?",
            "criteria": {
                "true": "Explicitly time-sensitive",
                "false": "No urgency expressed"
            }
        },
        "department": {
            "type": "choice",
            "instructions": "Which team should handle this?",
            "criteria": {
                "billing": "Payments, invoicing, refunds",
                "technical": "Bugs, outages, integrations",
                "sales": "Pricing, upgrades, new accounts"
            }
        },
        "frustration": {
            "type": "score",
            "instructions": "How frustrated is the customer?",
            "criteria": ["Calm", "Frustrated", "Very angry"]
        }
    }
}

response = requests.post(url, json=payload, headers=headers)
result = response.json()

print(result["answers"]["department"]["choice"])       # "billing"
print(result["answers"]["is_urgent"]["noul"])          # 0.92
print(result["answers"]["frustration"]["score"])       # 1.6
JavaScript / TypeScript — Quick Start
// Using fetch API
const url = "https://tokenra.io/v1/decisions";

const response = await fetch(url, {
    method: "POST",
    headers: {
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json"
    },
    body: JSON.stringify({
        model: "jev-latest",
        state: "Help! My payouts have been failing for 3 days.",
        questions: {
            is_urgent: {
                type: "noul",
                instructions: "Does this message convey urgency?",
                criteria: {
                    "true": "Explicitly time-sensitive",
                    "false": "No urgency expressed"
                }
            },
            department: {
                type: "choice",
                instructions: "Which team should handle this?",
                criteria: {
                    billing: "Payments, invoicing, refunds",
                    technical: "Bugs, outages, integrations",
                    sales: "Pricing, upgrades, new accounts"
                }
            },
            frustration: {
                type: "score",
                instructions: "How frustrated is the customer?",
                criteria: ["Calm", "Frustrated", "Very angry"]
            }
        }
    })
});

const result = await response.json();

console.log(result.answers.department.choice); // "billing"
console.log(result.answers.is_urgent.noul);    // 0.92
console.log(result.answers.frustration.score); // 1.6

The three primitives

The entire API is three question types. That's not a limitation — it's the design.

✅

Choice

Picks one option from a set you define. Returns the chosen option, full probability distribution over all options, and confidence. Up to 255 options.

📊

Score

Rates state along an ordered rubric (2–10 levels). Returns a probability-weighted position that can land between levels, plus per-level probabilities and confidence.

🎯

Noul

Yes/no as a probability. Returns a number from 0 (no) to 1 (yes). No separate confidence field — the number is the belief.

Patterns that change your architecture

Confidence-gated routing

Jev is trained with RLCD (Reinforcement Learning for Calibrated Decisions), so confidence is meaningful in aggregate: higher confidence = higher accuracy. Set different thresholds per action based on what being wrong costs.

Python — confidence-gated routing
action = response.answers["intent"]

if action.confidence < 0.5:
    route_to_human(user_message)   # genuinely unsure
elif action.choice == "check_balance":
    show_balance(account_id)        # read-only, low bar
elif action.choice == "approve_transfer":
    if action.confidence > 0.85:   # moves money, high bar
        approve_transfer(account_id)
    else:
        ask_user_to_confirm("Approve this transfer?")

Speculative fan-out

Questions are evaluated in parallel — a tenth question costs tokens but almost zero extra time. Ask everything up front and let code decide what mattered. TypeSafe's cookbook: 13-question batching is 12.2× cheaper and 10.0× faster than asking one at a time, with identical answers.

SDKs & integrations

🐍 TypeSafe Python SDK → 📦 TypeSafe JS/TS SDK → 🔀 OpenRouter → ▲ Vercel AI Gateway → ◎ Netlify AI Gateway → 🔧 LiteLLM →

How to get an API key

Get your API key at tokenra.io. Once you have your key, export it:

Shell
export TYPESAFE_API_KEY="sk-..."

You can also access Jev through OpenRouter, Vercel AI Gateway, Netlify AI Gateway, or AIMLAPI — all listed above.

Frequently asked questions

How do I get a Jev API key?

Get your API key at tokenra.io, then export it as TYPESAFE_API_KEY. Tokenra provides immediate access to Jev API without waitlist.

You can also access Jev through OpenRouter, Vercel AI Gateway, Netlify AI Gateway, and AIMLAPI — those keys work today as well.

Is the Jev API free?

No, but output tokens are. Input is $0.042 per million tokens and output is billed at $0 — Jev returns typed decisions rather than generated text, so there is no output token stream to charge for.

In practice that means a typical classification or routing call costs a small fraction of a cent. Gateway resellers may add their own margin, so check their pricing page for the exact rate.

What's the Jev API endpoint?

A single endpoint: POST https://tokenra.io/v1/decisions. Authenticate with Authorization: Bearer <API_KEY> and send Content-Type: application/json.

The body carries your program state plus a questions map built from the three primitives — Choice, Score, and Noul. All questions are evaluated in one parallel pass and come back as one typed answer each. That's the entire API surface.

Jev API vs OpenAI API — which is faster?

Jev, for decision workloads. It answers in 70–500ms end to end, against 3s–329s for frontier LLMs on comparable structured tasks, because it scores a fixed set of typed options in a single pass instead of generating text token by token.

That advantage only applies to decisions. Jev cannot write prose, summarize, or call tools — if you need generated text, you still want an LLM. The common pattern is both: Jev on the hot path for routing, classification, and gating, an LLM behind it for the work that needs language.

More resources

Jev API docs — the full guide on this site: authentication, pricing per 1M tokens, and a copy‑paste request example.

Jev API Key guide — get, configure, verify, and rotate your key securely.

Official API reference — full HTTP spec, request/response shapes, error codes, rate limits.

TypeSafe Documentation — concepts, foundations, patterns, demos, cookbooks.

Introducing System One Models & Jev — the official launch post.