TypeSafe AI's decision model API that returns typed verdicts — choices, scores, or yes/no probabilities — instead of generated text.
TypeSafe AI homepage
1 / 2Overview
What is Jev?
Intelligence beyond chat.
What it is
Jev is TypeSafe AI's flagship "System One" decision model: the first public model built to decide rather than to write. You send a state (text or structured context) plus typed questions to POST https://api.typesafe.ai/v1/systemone, and it returns typed verdicts — a choice from your options, a score on your rubric, or a yes/no probability — each with a calibrated probability distribution and confidence value. No text is generated, so there is nothing to parse.
Why it matters
The classic pattern for classification work is to prompt a chat model and parse its output back into a label — slow, expensive, and fragile. Jev answers natively in the shape your code needs: one request can carry many questions evaluated in parallel against the same state, at a listed $0.042 per million input tokens with free output and 70–500ms end-to-end latency (TypeSafe's figures). OpenAI's Decisions API, launched October 6, 2026, is widely read as a direct answer to Jev.
Access and company
Jev 1.13 is a closed, managed API (no open weights) available in early access via a waitlist and through gateways including Vercel AI Gateway, OpenRouter, and Cloudflare. TypeSafe AI announced an $870M raise at a $7.5B valuation on October 9, 2026, led by Andreessen Horowitz.
Platforms and languages
Jev Availability
Platforms
Languages
Capabilities
Jev Key Features
Typed decision primitives
Three primitives — Choice (pick from up to 255 options), Score (rate against a rubric), Noul (yes/no as a probability) — each returning probabilities and confidence.
Decisions, not strings
No text generation and no parsing: answers arrive as values your code can switch on, sort by, or route with.
Parallel questions
Ask many typed questions about the same state in one call; each is evaluated independently and in parallel.
Low per-decision cost
$0.042 per million input tokens, output free — a fraction of a chat-model call for classification workloads.
Fast responses
TypeSafe quotes 70–500ms end-to-end, with a 64k-token request budget for state plus questions.
SDKs and agent skill
Python and JavaScript/TypeScript SDKs, a plain HTTP endpoint, plus an agent skill for coding assistants like Claude Code, Cursor, and Codex.
Language guidance
English is the primary training language; other languages work but with lower accuracy — test with your own content.
Plans and access
Jev Pricing
Paid
Common questions
Jev FAQs
How do I get access to Jev?
TypeSafe is opening early access and onboarding developers from the waitlist; you can also reach Jev through the Vercel AI Gateway, OpenRouter, or Cloudflare. API keys are created in the TypeSafe console.
What does Jev cost?
$42 per billion input tokens — $0.042 per million — and output is free, since no output tokens are generated. Output tokens are free; only input is billed.
How is Jev different from an LLM with structured outputs?
A chat model with JSON mode is a text-generation system coerced into producing a decision you then parse. Jev skips generation entirely and returns typed values with probabilities and confidence that your code can branch on directly.
Can Jev write essays or code?
No. Jev does not generate open-ended text — it judges drafts, picks tools, or selects models, while a generative model does the writing. Its output space is limited to the options you define.
Is my data used for training?
No. TypeSafe does not train on customer requests or responses, and enterprise customers can opt into zero data retention (ZDR).
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
