Classification and routing
Teams that need to assign tickets, messages, documents or requests to a known category and route each result to code, a specialist model or a person.
Independent tool overview
Jev AI: fast, typed decisions for software
Jev is TypeSafe AI's first System One model, built to turn text or structured state into fast, typed decisions that software can use directly.
Visit the official Jev site ↗
- Best for
- Classification, routing, scoring, verification and guardrails
- Current model
- jev-1.13.0, available through jev-latest
- Input
- Text, JSON objects or arrays of text values
- API price
- $0.042 per million input tokens; output is free
Overview
What Jev is
Jev is designed for decisions inside software rather than open-ended chat. An application sends text or structured state plus one or more typed questions, and the model returns bounded values and probability distributions instead of generating a paragraph that must be parsed.
Its three question types cover choosing from known options, scoring against an ordered rubric and estimating whether a statement is true. Multiple questions can share the same state and run in parallel, while application code keeps control of thresholds, branching, arithmetic and escalation to a person or a more general model.
Use cases
Who Jev is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Confidence-gated automation
Workflows that should act automatically above a chosen confidence threshold and escalate uncertain cases for review.
Scoring and verification
Applications that evaluate content against an ordered rubric, check a claim or citation, rank candidates, or add guardrails around another model.
High-volume software decisions
Developers making many narrow, repeated judgments where typed output, low per-call cost and parallel evaluation matter more than generated prose.
Capabilities
Core Jev features
Choice questions
Selects one option from a developer-defined set and returns the selected value, a probability for every option and a confidence score.
Score questions
Rates the supplied state against ordered descriptive levels and returns a score, the level probabilities and confidence.
Noul questions
Evaluates a yes-or-no statement and returns a probability from zero to one, which code can compare with its own action threshold.
Parallel questions
Evaluates multiple independent questions against the same state in one request, including a mix of Choice, Score and Noul types.
Typed API responses
Keeps the possible output structure defined in advance so applications can branch on returned values without parsing free-form generated text.
Versioned models and aliases
Supports the stable jev-latest alias for convenience and versioned model IDs when a team needs to pin behavior and retest thresholds before upgrading.
Python and JavaScript SDKs
Provides official client SDKs alongside the HTTP API, including retries with backoff and model-listing support.
Process
How the Jev workflow works
Step 1
Prepare the relevant state
Send only the text or structured fields needed for the decision; filter unrelated context in code before the request when possible.
Step 2
Define atomic questions
Choose Choice, Score or Noul for each narrow judgment and spell out the available options, rubric levels or exact condition.
Step 3
Call the System One endpoint
Submit the shared state and questions to POST /v1/systemone with jev-latest or a pinned version such as jev-1.13.0.
Step 4
Act on values and confidence
Use normal code to apply thresholds, combine dimensions, perform calculations, log the answering model and route uncertain cases for review.
Cost
Jev pricing and free plan
TypeSafe charges for Jev input tokens and does not charge for output tokens. The published standard limits are high, but TypeSafe says those limits can change while the service scales.
Jev 1.13 API
$0.042 per million input tokens
Standard usage of the current Jev model through the System One API.
- Output tokens are free
- Published standard limits are 250,000 tokens per second and 1,200 requests per minute
- The current model accepts up to 64k total tokens per request, with a 32k limit for state plus the longest question
Custom and enterprise
Contact TypeSafe
For organizations that need higher limits or enterprise data-handling options.
- Higher rate limits are available
- Enterprise customers can discuss zero data retention
- Contact sales@typesafe.ai for current terms
Pricing checked . Check current pricing at the source ↗
Assessment
Jev strengths and limitations
Where it stands out
- Returns bounded, typed values that application code can consume directly instead of parsing generated prose.
- Runs many independent questions against one shared state in parallel, with little additional response time per question.
- Publishes probabilities and confidence for decisions, giving developers a practical way to gate automation and escalate uncertainty.
- The current per-token price and free output make narrow, repeated decisions inexpensive compared with general-purpose text generation.
What to consider
- Jev is not trained to write text, summarize, code or produce open-ended answers; use a generative model when the output must be prose.
- The current model accepts text only, so images, audio, video and binary inputs must be converted into relevant text or structured fields first.
- Jev 1.13 is weak at counting, exact arithmetic and date comparisons; keep calculations and deterministic rules in code.
- Literal wording, indirect instructions, contradictory criteria and large amounts of irrelevant state can reduce accuracy.
- English is the primary training language, and TypeSafe recommends testing non-English workloads on representative content before relying on them.
- Adversarial text in the state can influence a result, so production integrations still need careful prompts, thresholds and edge-case testing.
Compare
Jev alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Consumer
GPT-5.4 mini & nano
Choose GPT-5.4 mini or nano when you need a broader generative model that can write, reason and handle tasks beyond a fixed decision space.
Explore GPT-5.4 mini & nano →Consumer
Gemini 3.1 Flash-Lite
Choose Gemini 3.1 Flash-Lite when low-cost general-purpose generation and multimodal model capabilities matter more than Jev's bounded typed outputs.
Explore Gemini 3.1 Flash-Lite →Coding
Ollama
Choose Ollama when running models locally or controlling the hosting environment is more important than using TypeSafe's managed decision API.
Explore Ollama →Questions
Jev FAQs
What is Jev AI?
Jev is TypeSafe AI's first public System One model. It takes text or structured state plus typed questions and returns decisions, probabilities and confidence values that software can use directly.
Is Jev a chatbot?
No. Jev is designed for bounded decisions inside software, not conversation or open-ended text generation. It does not write prose responses.
What does Jev return?
Jev supports Choice questions that select from known options, Score questions that evaluate ordered rubric levels and Noul questions that return a yes probability. Choice and Score also return confidence.
How much does Jev cost?
As of September 22, 2026, TypeSafe lists Jev 1.13 at $0.042 per million input tokens. Output tokens are free. Custom and enterprise arrangements are available from TypeSafe.
How much context can Jev use?
The current Jev 1.13 documentation lists a 64k-token total request limit and a 32k-token limit for the state plus the single longest question. Accuracy can still fall when the state contains a lot of irrelevant detail.
Can Jev process images, audio or video?
Not directly. Jev 1.13 accepts text, a JSON object or an array of text values. Non-text inputs must first be converted into relevant text or structured fields.
When should you not use Jev?
Use another model or ordinary code when you need generated text, complex multi-step reasoning, exact math, counting or date calculations. Jev is strongest when the answer space is bounded and the decision can be expressed as one atomic question.
Bottom line
Our Jev verdict
Jev is a strong fit for developers who need fast, inexpensive classification, routing, scoring or verification inside a controlled software workflow. Its value comes from typed decisions, parallel questions and confidence-aware branching, not from replacing a chatbot or reasoning model. Teams should keep deterministic work in code, test thresholds on their own data and route uncertain cases for review.
Visit Jev website ↗Similar Miscellaneous Tools
View all Miscellaneous AI tools →
Canto
WisprFlow’s new model for accurate dictation in loud places
Eleven v4
ElevenLabs' new top-ranked voice model, with a faster Turbo version
Mercury Books
Mercury's built-in bookkeeping with an AI assistant for financial questions
Mercury Voice
Inception's cheap, speedy reasoning model for AI voice agents
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.