TypeSafe AI · 2026-09-15 · major
Jev — TypeSafe's model returns typed decisions, not text
Jev is the first System One Model from TypeSafe AI. It takes unstructured state plus typed questions and returns typed answers with a confidence score, priced at $0.042 per million input tokens with output tokens free.
A model that answers typed questions with calibrated probabilities instead of writing prose you have to parse.
Quick facts
| Maker | TypeSafe AI |
|---|---|
| Model | Jev (first System One Model) |
| Training method | RLCD (Reinforcement Learning for Calibrated Decisions) |
| Price (input) | $0.042 / 1M tokens |
| Price (output) | Free |
| Latency | 70–500 ms |
| Availability | Early access — Python and JavaScript SDKs plus an HTTP API |
Pricing
| Input · quoted as $42 per billion tokens | $0.042 / 1M tokens |
|---|---|
| Output · output tokens are not charged | $0 / 1M tokens |
What is it?
Jev is the first System One Model from TypeSafe AI: you send it unstructured state plus typed questions, and it returns typed answers your code can use directly. Rather than free text, Jev answers three question shapes — Choice (pick one of your options), Score (rate against described levels), and Noul (a yes/no probability). Every answer comes back with a calibrated confidence number, so the calling code can decide when to escalate to a human.
How does it work?
Training replaced RLHF with a method TypeSafe calls Reinforcement Learning for Calibrated Decisions (RLCD), paired with a parallel sampler and a stack tuned for automation rather than conversation. Because no answer is written out token by token, many questions can be answered at once, and replies land in 70 to 500 milliseconds. TypeSafe reports that this is 40 to 200 times faster than frontier LLMs on the same System One tasks.
Why does it matter?
Wiring an LLM into automation usually means coaxing free text into a schema and then guessing how sure the model was. TypeSafe prices Jev at $0.042 per million input tokens and charges nothing for output tokens, so triage, routing and review steps that were too expensive to run on every record become cheap enough to run on all of them. Founder Diogo Almeida previously worked at OpenAI on the research behind ChatGPT.
Who is it for?
backend and automation engineers
Frequently asked questions
- How much does Jev cost?
- Jev is priced at $0.042 per million input tokens, which TypeSafe also writes as $42 per billion tokens, and output tokens are not charged at all. That pricing only covers the input state and questions you send, so the cost of a decision scales with how much context you hand Jev rather than with how much it replies.
- How is Jev different from asking an LLM to reply in JSON?
- Asking an LLM for JSON still produces text that can be malformed or drift from your schema, and gives no reliable measure of certainty. Jev instead answers typed primitives — Choice, Score, and Noul — so the returned value already matches the type your code expects, and each answer carries a calibrated probability you can threshold on before acting.
- What kinds of tasks is Jev built for?
- TypeSafe's public evaluations run Jev on four workflow tasks: triaging security alerts into close, escalate or contain; reviewing support-agent traces to decide whether a human should look; deciding whether a vendor invoice is paid, held or returned; and choosing how a customer-service assistant should respond. All four are decisions software acts on, not text a person reads.
- Can I call Jev from a coding agent?
- TypeSafe publishes an agent-skill page in its documentation covering Claude Code integration, alongside Python and JavaScript SDKs and a plain HTTP API. Access uses an API key from the TypeSafe console, set through the TYPESAFE_API_KEY environment variable, and the default model id is jev-latest.
Try it
pip install typesafe-sdk — then call client.system_one(...) against model jev-latest