TypeSafe AI has released Jev, a system-one model designed to return typed decisions with probabilities rather than generating text. The tool operates as a transformer-based engine that accepts state and typed questions, outputting structured data for code to branch on. It is currently available as a hosted API in early access behind a waitlist. TypeSafe has not published model weights, parameter counts, or options for self-hosting.
What is a System One Model?
The name references Daniel Kahneman’s distinction between fast intuition and slow reasoning. TypeSafe argues that Reinforcement Learning from Human Feedback (RLHF) tuned for human preference created chat models prone to overconfidence and mode dropping. Those flaws usually require a human in the loop.
Jev employs a new stack combining a novel architecture, a parallel sampler, and Reinforcement Learning for Calibrated Decisions (RLCD). The company has not disclosed the specific architecture details.
How the Jev API Works
A single endpoint handles all requests: POST https://api.typesafe.ai/v1/systemone. The request body contains a state, a model, and a map of questions. Documentation defines three question types.
| Primitive | Asks | Returns |
|---|---|---|
| Choice | Pick 1 option from a list | choice, probabilities, confidence |
| Score | Rate against ordered levels | score, probabilities, confidence |
| Noul | Is this statement true? | noul, a probability from 0 to 1 |
Questions run in parallel and in isolation against the same state. TypeSafe states that adding questions barely changes response time. A Choice supports up to 255 options.
Developers install the Python SDK using pip install typesafe-sdk for Python 3.10 or later. A JavaScript SDK is available as @typesafe-ai/sdk. The quickstart guide also covers cURL and an agent skill for Claude Code.
Confidence is the Product
Every Choice and Score answer carries a confidence value from 0 to 1. TypeSafe derives this from the shape of the probability distribution. In the documentation example, billing wins at 0.84. Confidence is only 0.596 because technical still holds 0.159.
The documentation suggests three paths: act on high confidence, review the middle range, and send low confidence cases to a human. Thresholds should scale with the cost of a wrong action.
Pricing, Speed, and the Benchmark Fine Print
Jev costs $42 per billion input tokens. TypeSafe quotes existing LLMs at $0.20 to $10 per 1 million input tokens. In a recorded demo, Jev finished in 0.114 seconds for $0.000081. GPT-5.6 Terra took 8.566 seconds for $0.013880.
The TypeSafe team claims it is 193.6 times faster and 444.6 times cheaper. Those figures come from TypeSafe’s own workflow evals. However, several caveats apply:
- The reference answer is the average of GPT-6 Astra and Fable 5.1.
- TypeSafe’s capabilities team wrote the workflows.
- TypeSafe expects these gains to sit at the high end of real use.
- TypeSafe says it cannot prove the price is unsubsidized.
‘Zero hallucinations’ means schema matching is guaranteed. The 0% figure is not empirical. Answers can still be wrong.
What Developers are Building with Jev
Community projects appeared within days of launch. Here are some examples:
- Command safety: Vercel CEO Guillermo Rauch reported Jev up to 18 times faster at p95 than GPT Luna, and more accurate. His post said the fx reviewer still ran on Luna. Engineer Pranit Sharma shared the benchmark.
- Email triage: Bryo AI CTO Nikhil Mudholkar found Gemini slightly more accurate, but 10 to 20 times more expensive.
- Browser agents: Browser Use’s jev-ultrafast ran a Zürich to London Google Flights search in 7.1 seconds.
- Phone agents: Droidrun’s mobile-jev drove Uber on a real Android phone: 9 actions in about 21 seconds. No booking was completed.
- Video scoring: jevmeter scores every sentence of a debate for about $0.05.
- Live typing: Steve Krouse’s Typewriter updates 16 judgments as you type.
- Games: Jev completed StarCraft’s first combat mission. It also runs the guards in heist-one.
- Agent guardrails: jev-guard rates each tool call as deny, ask, or allow.
- Data and homes: pg-jev adds plain-language filters to Postgres. HA-Jev turns answers into Home Assistant entities.
What it means
For people writing code, this changes the interaction from prompting for text to querying a database of logic. You provide the context and ask specific questions. The system returns structured answers with confidence scores. You can then branch your code based on those scores without needing to parse natural language. This removes the need for manual validation of every output string.
