What is a System One model, and what is Jev?

System One models are built to make decisions inside software rather than produce text for people to read. Jev is TypeSafe's first public System One model: send it a state plus structured questions and it returns typed decisions with probabilities and confidence that your code can act on.

How is Jev different from an LLM's JSON or structured-output mode?

Jev does not generate strings at all. The possible outputs and their structure are declared up front, so the model cannot return a type error, and every answer arrives with a calibrated probability instead of text that has to be parsed and validated.

Input is billed at $42 per billion tokens, or $0.042 per million, and output tokens are not charged. Higher rate limits are offered on custom and enterprise plans through sales@typesafe.ai.

How fast is Jev, and how are the speed claims measured?

TypeSafe publishes end-to-end response times of 70ms to 500ms. The 193.6x faster and 444.6x cheaper figures on the homepage come from its own workflow evals, which compare models against the average predictions of GPT-6 Astra and Fable 5.1.

What are Jev's input and rate limits?

Jev accepts text only, as a string, JSON object or array. A request carries up to 64k tokens in total, with 32k for the state plus the longest single question, under limits of 250,000 tokens per second and 1,200 requests per minute.

Jev is in early access behind a waitlist on typesafe.ai. Once you are in, you take an API key from the console and call POST /v1/systemone directly or through the Python or JavaScript SDK.

TypeSafe AI

Visit website

An AI lab whose System One model Jev returns typed decisions with calibrated confidence, not text.

PaidAI API AI Developer Tools AI Workflow

Are you the maker of TypeSafe AI?Claim tool Advertise

Visit website

What is TypeSafe AI?

TypeSafe AI is a San Francisco AI lab that trains models for software instead of chat. Its first public model, Jev, takes a block of state plus typed questions and returns choices, scores and truth values with calibrated probabilities, so code can branch on the result without parsing text.

What can TypeSafe AI do?

Typed decisions
Returns choices, scores and truth values instead of generated text.
Calibrated confidence
Every answer carries a probability distribution and a confidence score.
Three question primitives
Choice, Score and Noul can be mixed in a single API call.
Parallel evaluation
All questions run at once against one shared state per request.
Sub-second responses
Published end-to-end latency of 70ms to 500ms per call.
Token-based pricing
$0.042 per million input tokens; output tokens are not billed.
REST API and SDKs
POST /v1/systemone with official Python and JavaScript clients.

Who is TypeSafe AI for?

AI automation software: Code keeps control of the flow while Jev handles the semantic judgements, so a workflow can run in the background millions of times without a person in the loop.
Search and retrieval: Teams replace or supplement embeddings in a RAG pipeline by scoring query-to-candidate relevance and reranking results with pairwise comparisons.
LLM guardrails: Semantic checks sit on every LLM input, output and tool call to catch jailbreaks, prompt injection and policy violations at a fraction of the LLM's own cost.
Recruiters and hiring managers: Resumes, applications and interview feedback are scored against explicit job-related criteria, and candidates are matched to roles and routed onward.
Scientific discovery: Researchers screen papers against inclusion and exclusion criteria for systematic reviews and check whether cited passages actually support a claim.