Awesome Jev GuideShould you use Jev, and how?

What is Jev? ​

It does not generate text. You send a state plus typed questions; it returns structured decisions with probabilities that your code can branch on directly. Three question types, from the official docs:

Jev was released 2026-09-15 by TypeSafe AI (launch post). Headline figures from the official model docs, retrieved 2026-09-20: 70–500 ms latency, $0.042 per 1M input tokens, output free, up to 255 questions per call.

Should you use Jev? ​

A short decision tree. One lesson shows up across the independent evals listed on the research page: asking Jev a single big vague question works poorly — the reliable pattern is to decompose the task into atomic questions and gate actions on confidence.

Quickstart ​

This example follows the official TypeSafe docs. The commented outputs are from our own run against the live API on 2026-09-20.

from typesafe_sdk import Choice, Score, Noul, TypeSafeClient

client = TypeSafeClient()

response = client.system_one(
    state="Hi, my Stripe integration keeps failing for 3 days. Help ASAP.",
    questions={
        "department": Choice(
            instructions="Which team should handle this",
            criteria={"billing": "Payment issues",
                      "technical": "Bugs or integrations",
                      "sales": "Pricing"},
        ),
        "frustration": Score(instructions="How frustrated the customer appears",
                             criteria=["Calm", "Frustrated but civil", "Very angry"]),
        "is_urgent": Noul(instructions="The message conveys urgency"),
    },
)

print(response.answers["department"].choice)      # "technical"
print(response.answers["department"].confidence)  # 0.75
print(response.answers["is_urgent"].noul)         # 0.99

TYPESAFE_API_KEY=ts-your-key uv run --with typesafe-sdk triage.py

Cost and latency ​

Figures below are from the official docs (retrieved 2026-09-20). The "193× faster / 444× cheaper" comparison in TypeSafe's launch post is self-measured; independent evaluations that confirm or qualify these claims are collected on the research page.

Cost formula: monthly $ = daily calls × avg state tokens × 30 × 0.042 / 1M — e.g. 100k calls/day × 500 tokens → 1.5B tokens/month → $63/month.