Jev 1.13: the current public model
As of the public models page, the current versioned Jev model is jev-1.13.0. The aliases jev-latest and jev-preview both point at that ID. There is no separate preview build until TypeSafe ships one. The jaggedness note applies to jev-1.13 and was last reviewed 2026-09-17.
This page is for builders who already heard of Jev / System One and need a pin-and-character briefing. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm the models page on docs.typesafe.ai. We do not sell keys. We will not invent a 1.14 date.
Definition
Jev is TypeSafe’s first public System One model: unstructured state in, typed probabilistic decisions out. No text generation. jev-1.13.0 is the versioned id you should log and, if you tune thresholds, pin. jev-1.13 without .0 is the marketing line unless the models page lists it as an alias.
The response model field reports the versioned ID that actually answered, even if you sent an alias. Log it.
Published envelope (vendor claims — confirm there)
- Price: $42 / billion input tokens, $0.042 / million. Output tokens are free.
- Rate limits: 250,000 tokens per second and 1,200 requests per minute, described as dynamically adjusting.
- Context: 64k tokens for state plus all questions; 32k for state plus the longest question.
- Input: text only.
- Language: English primary; other languages accepted with currently lower accuracy.
Those are vendor-published figures. Homepage multiples (193.6× / 444.6×) are attributed to workflow evals and described as the higher end of real-world gains. Latency 70ms–500ms and “40x–200x” are the same class of claim. This site does not add independent benches.
Mental model: pin when floors are fitted
If you have tuned confidence thresholds against a specific version, TypeSafe tells you to pin that versioned ID and move on your schedule. Aliases move when a release ships. GET /v1/models currently lists aliases. Versioned IDs are still accepted on POST /v1/systemone whether or not they appear in the list.
Schema-safe ≠ correct on every version. “Can’t hallucinate” remains schema-constrained, not factually right.
Character of 1.13 (jaggedness)
TypeSafe’s own jaggedness summary: fast, calibrated, good at common-sense judgment; weaker on extra indirection, literal about wording, and unreliable at numeric precision. Details live on jaggedness.
Practical consequences:
- Keep arithmetic in code. Jev is not a calculator.
- Filter state. Accuracy falls as unrelated detail grows.
- Do not force generation by chaining Choices — slow and weak (why Jev gives up generation).
- Ask atomic questions. Literal wording means the missing half of the instruction is missing from the judgment.
Training name: RLCD. No full public paper — do not invent datasets or losses for 1.13.
See also
FAQ
Is jev-1.13 (no .0) valid?
Treat jev-1.13 as the marketing line unless the models page lists it as an alias.
When will 1.14 ship?
We will not invent a date. Watch the models page.
Are homepage speed multiples about 1.13?
TypeSafe attributes workflow-eval multiples to that work. Label them vendor-reported.
Does 1.13 remember prior POSTs?
No hidden session. Put the history slice you need in this request’s state.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.