What is jev?

Jev processed 37 documents with 21 questions each, returning 777 judgments in under 0.7 seconds for approximately a quarter of a cent.

“In less than 0.7 seconds, Mike said, Jev quote unquote read all 37 documents and answered all 21 questions for each, returning 777 judgments for an estimated quarter of a cent.”

Jev by Typesafe is a judgment model trained with RLCD (reinforcement learning for calibrated decisions), built by a co-inventor of ChatGPT, and positioned as a faster and cheaper alternative to LLMs for structured decision tasks.

“I've spent the last 2 years in stealth building a new way to train models, RLCD, or reinforcement learning for calibrated decisions and a new type of frontier AI model that we are releasing today, Jev.”

Jev works as a pure classifier: it receives an output schema with predefined categories and selects among them, returning type-safe structured data directly usable in code with no intermediate reasoning or text generation.

“It doesn't generate the categories itself. it looks at the categories that you've passed into it as like a model.”

Existing LLMs optimize for human preference in chat responses, while judgment models (system one models) optimize for calibrated decisions with epistemically honest probabilities.

“Whereas existing LLMs optimize for human preference, i.e. write-ups and chat responses that human raiders prefer. The new system one models, the first of which is Jev, optimized for calibrated decisions or answers with epistemically honest probabilities.”

Jev is a decision-only AI model that outputs no text, returns type-safe structured data, and processes queries in roughly 200 milliseconds, positioning it as a fast classification layer distinct from conversational LLMs.

“What's really interesting, if you look at the actual spec of Jeb, it doesn't generate any text at all.”

Judgment models fill the gap between simple deterministic code and slow expensive LLMs by producing calibrated probabilities around specific questions rather than long text outputs.

“This allows a class of decision-making that was neither suited to dumb, unintelligent code nor to slow, expensive LLMs.”

Ryan Vogel on Jev's output model: no reasoning, no intermediate text, just the decision.

“From our point of view, it doesn't reason. It doesn't have any other text output. It just gives you the output.”

Jev processes any query in approximately 200 milliseconds, regardless of input or output structure.

“It takes around 200 milliseconds per query to Jev no matter like what the input output structure is.”

Nathaniel Whittemore on Jev acting as an instant quality-check layer for knowledge work.

“Typesafe's model is so fast at turning fuzzy tasks into clear structured answers that it could act as a kind of code llinter for knowledge work.”

Jev from Typesafe is 20 to 200 times faster and 40 to 400 times cheaper than comparable alternatives, with output tokens free.

“Jev is 20 to 200 times faster, 40 to 400 times cheaper with output tokens free.”