Jev AI: Pricing, API & How It Works

Jev AI is TypeSafe AI’s System One model that returns typed decisions with probabilities, not text, in 70-500 ms. See Jev pricing, the API and its limits.

Instead of writing an answer, Jev picks an option, scores something on a scale or gives a yes/no probability, so your code can branch on the result without parsing text. TypeSafe pitches it for the small, high-volume judgments that are too fuzzy for a hand-written rule but too small for a frontier LLM call.

What is Jev AI and why is it new?

TypeSafe AI announced Jev on September 15, 2026 as the first public System One model, a new class of AI built to make fast, structured decisions for software. Founder Diogo Almeida says he helped build the research behind ChatGPT at OpenAI before spending two years in stealth on a new architecture, a parallel sampler and a training method called Reinforcement Learning for Calibrated Decisions (RLCD). GIGAZINE covered the launch the next day.

Jev launched in early access. On September 27, 2026, TypeSafe’s homepage announced “No more waitlist”: the platform is now open to everyone. The name comes from economist William Stanley Jevons, because TypeSafe expects cheaper intelligence to unlock far more use cases.

How Jev works

You send one request with a state, which is the text or JSON you want judged, and a set of named questions. Jev evaluates every question in parallel against that state and returns typed answers, according to TypeSafe’s documentation. There are three question types:

• Choice: pick one of up to 255 labelled options, with a probability for each option and a confidence.
• Score: place the input on an ordered scale of 2 to 10 levels, with the full distribution and a confidence.
• Noul: a calibrated yes/no, returned as a probability from 0 to 1.

1. Create an account and get an API key in the TypeSafe console, or try queries in the Playground.
2. Send a POST request to api.typesafe.ai/v1/systemone with the model jev-latest, your state and your questions.
3. Branch in code on the typed answers, and use confidence to decide when to act and when to escalate to a person or a reasoning model.

Official Python and JavaScript SDKs are available, plus an agent skill for coding agents. If you’re weighing this against JSON mode in a normal LLM, our guide to connecting LLMs with structured data covers the usual approach.

Jev Pricing (2026)

Jev charges only for input tokens. Output tokens are free. We checked TypeSafe’s Models page on October 7, 2026.

Item	Jev 1.13 (jev-1.13.0)
Input tokens	$0.042 per million ($42 per billion)
Output tokens	Free
Rate limits	100,000 tokens per second and 80 requests per second
Context	64k tokens per request; 32k for the state plus the longest question
Higher limits and zero data retention	Custom and enterprise plans through TypeSafe sales

For scale, TypeSafe says a demo that played DOOM with about 10 Jev calls a second cost around $7 an hour. On its own workflow tests it claims Jev was 193.6 times faster and 444.6 times cheaper than LLMs, and says those figures are at the high end of real-world gains. TypeSafe also says it can’t prove its pricing isn’t subsidized, and expects prices to go down, not up.

Is it worth it? If you run lots of classification, routing or scoring calls through an LLM today, Jev is cheap enough to test on a real workload. It isn’t a replacement for a chat model.

Best use cases for Jev

• Ticket and email routing: send each message to the right team with a single Choice question.
• Moderation and guardrails: screen messages going into and out of an LLM app.
• Lead and risk scoring: turn fuzzy judgments into scores you can threshold.
• RAG filtering and re-ranking: decide which retrieved passages reach the answering model.
• Agent decisions: pick the next tool or skill for an agent, alongside frameworks like n8n.

Limitations

• No text output: Jev can’t write replies, code or explanations of its reasoning.
• Text input only: images, audio and video must be turned into text first.
• English first: other languages work, but accuracy is best in English.
• Shifting rate limits: TypeSafe says limits can change without notice as it adds capacity.
• Vendor benchmarks: speed and cost claims come from TypeSafe’s own tests.

Jev alternatives

• OpenAI SDK: structured outputs from GPT models when you also need generated text.
• Claude Sonnet 4.5: a general LLM that can return JSON, at higher cost per call.
• n8n: workflow automation where an LLM or Jev handles each decision step.
• Best AI models: our comparison of the top LLMs for text, coding and agents.

Jev is a genuinely different kind of AI model: instead of text, it returns typed decisions with probabilities, fast and at a fraction of a cent per call.

It won’t replace a chat model, and the impressive benchmarks are TypeSafe’s own. But if your app makes thousands of small routing or scoring calls through an LLM, Jev is cheap enough to test on real traffic.