Skip to content

TypeSafe AI ships Jev, a cheap decision layer built to replace LLM calls in routing and triage

Short answer

TypeSafe AI launched Jev, a non-LLM decision model that answers yes/no, scoring or multiple-choice questions with a confidence level in 70-500ms, costing $0.042 per million input tokens with free output. It targets routing, classification and screening steps — like support ticket triage or CV screening — where LLM calls are too slow or expensive, though its accuracy sits at 67.8%.

What this means for operators

For a support or ops team that currently pays an LLM to tag ticket urgency, screen resumes or decide which model handles a request, Jev is a much cheaper intermediate step: it returns a structured score or choice with a confidence value instead of free text, so you can auto-route high-confidence cases and send low-confidence ones to a human. The catch is the 67.8% accuracy figure and the lack of any explanation for how it reached a decision, so it's a triage layer, not a replacement for judgment on anything with real downside if it's wrong — and adversarial inputs (a crafted support message, a doctored CV) can still hijack its reasoning without strong guardrails around it.

TypeSafe AI has launched Jev, a decision-making AI model the company frames as a "System One" model — fast and instinctive rather than slow and deliberate, borrowing the terms from Daniel Kahneman. Unlike a large language model, Jev doesn't generate text, can't chat, and according to TypeSafe cannot hallucinate in the structural sense: it always returns output in the same typed format, even if the content of that answer is wrong.

Jev works with three question types — Noul (yes/no, scored 0 to 1), Score (rated against a rubric), and Choice (pick one option from a set) — applied against a "state," whatever text you're asking about. Each answer comes with a confidence level. In a CV-screening example run by Zapier, Jev scored a candidate's LLM experience, technical depth and career-progression pattern, each with its own confidence score, from 78% to 100%.

On speed and cost, TypeSafe reports Jev returns a decision in 70-500ms regardless of how many questions are attached, at $0.042 per million input tokens with output priced at effectively zero. In a WikiRace demo navigating from the Wikipedia page for Baseball to Sun using only on-page links, Jev completed the task in 0.419 seconds for $0.047, versus Claude Sonnet 5 (3.724 seconds, $3.31), Claude Haiku 4.5 (4.975 seconds, $1.03) and GPT-5.6 Terra (9.453 seconds, $2.04 cents, with one hallucination).

Suggested uses include routing requests to a cheaper or stronger LLM before inference, screening support messages for topic, urgency and sentiment, labeling large batches of documents or proposals, checking AI tool outputs for harmful or low-quality calls, and gating actions in a network based on a safety judgment.

TypeSafe discloses Jev's accuracy at 67.8%. The model lacks multi-step reasoning, can't explain how it reached a conclusion, only processes text (audio, image and video must be converted to descriptions first), and remains vulnerable to adversarial prompts — an area TypeSafe says it is still working on.

Jev is available now through TypeSafe's API and playground, with $5 in free monthly credits. There's no native Zapier integration yet; Zapier's writeup describes wiring it in through API by Zapier and routing the output with Filter by Zapier.

Source: Zapier Blog · In the Atlas: Zapier →

Next step

Visibility Analyzer

This is what the Visibility Analyzer measures on a real site: which answers cite you, which pages an engine cannot retrieve, and what to fix first. Free to run.

Run a free visibility audit

Free to run. No card.

Fee
Free
Length
One run, minutes

Free tier: two analyses a day, no card required.