Article
OfficialJev Guides & Articles
Launch: Introducing System One Models & Jev
The launch story: why TypeSafe built decision models, the side-by-side demo, the Doom and Wikiracing bots, and the published workflow evals.
- Format
- Article
- Reading time
- 2 min
- Published by
- TypeSafe AI
- Last verified
- Sep 2026
Used inJev for Gaming
Highlights
- Jev launched in early access on Sep 15, 2026, trained with Reinforcement Learning for Calibrated Decisions (RLCD) and optimized for structured outputs.
- Published pricing is $0.042 per MTok of input with output tokens free, and end-to-end response times of 70ms-500ms.
- Workflow evals use the average of Astra and Fable as reference probabilities, with four workflows published at evals.typesafe.ai.
- The Doom bot issued about 10 queries a second at roughly $7/hour; Wikiracing picks among hundreds to thousands of links per step.
- Jev supports cardinality up to 255; higher-cardinality choices use a two-stage score-then-choose system, which occasionally slows it down.
Watch out
The post itself flags the evals as run from West Coast laptops and cannot prove pricing is not subsidized, and Jev was early access at launch, so read the 193.6x/444.6x multiples as directional.
Questions about this entry
Short answers drawn from the linked project or page.
Published pricing of $0.042 per million input tokens with output free, and end-to-end response times of 70ms-500ms. The 193.6x and 444.6x multiples come from the workflow evals and should be read as directional.
They use the average of Astra and Fable as reference probabilities, with four workflows published at evals.typesafe.ai. The post itself flags the evals as run from West Coast laptops and cannot prove pricing is not subsidized.
Reinforcement Learning for Calibrated Decisions, or RLCD, paired with a new model architecture and a parallel sampler for efficient inference. The launch post frames this stack as the reason Jev can make calibrated structured decisions.
Jev launched in early access on Sep 15, 2026, the same day as the announcement. The post itself notes it was early access, so the 193.6x and 444.6x multiples should be read as directional rather than settled.
The Doom bot issued about 10 queries a second at roughly $7 per hour, and the Wikiracing bot chooses among hundreds to thousands of links per step in a wiki-racing game, illustrating high-frequency decisions.
Cardinality is supported up to 255. Higher-cardinality choices use a two-stage score-then-choose system, which the post says occasionally slows it down.
No. Jev gives up string generation and returns typed probabilistic decisions, so the post says it cannot hallucinate. It is described as a frontier-intelligence function call: unstructured state in, typed decisions out.
More Jev resources like this
TypeSafe's 11 September 2026 essay against benchmaxxing: a public eval can select the model even when nobody trains on that test.
Jev Guides & ArticlesArticle#official#benchmarks#evaluation
Official
The index of TypeSafe's worked examples, from parallel questions and reranking to guardrails, date extraction, and self-consistency, each with datasets and measured results.
Jev Guides & ArticlesDocs#official#docs#cookbook
Official
An open 0.6B replica of Jev that turns states and questions into full probability distributions without decoding answer tokens, trained and evaluated on Maze, Snake, and ViZDoom.
Jev Repos & SDKs#community#python#open-models
Community
TianyuCodings/NanoJevPython
TianyuCodings/NanoJevPythonBraintrust's provider page covers two ways to use TypeSafe Jev: instrumenting System One calls as decision spans, and running Jev as an LLM judge that turns typed choices into scores or labels.
Jev Guides & ArticlesArticle#community#press#explainer
Community
Langfuse's guide to decision-model evaluators runs OpenAI's gpt-6-luna or TypeSafe Jev over observations, where each Choice, Score or Yes/no question becomes one Langfuse score.
Jev Guides & ArticlesArticle#community#press#explainer
Community
The desk's index of every Jev measurement it has published, keeping each figure's original caveat: JevBench boards, Vercel gateway reach, Celeris and Artificial Analysis results, and the desk's own Jev Day probes which it says it did not rerun.
Jev Guides & ArticlesArticle#community#press#calibration
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.