Jev 1.13

Evaluates a state against typed questions and returns a choice, a score, or a yes-probability with calibrated probabilities, rather than text for a person to read. TypeSafe calls this class a System One model, and Jev is its first.

Input

$0.04

per 1M tokens

Output

Free

per 1M tokens

Context

64K tokens

Features

[choice-primitive][score-primitive][noul-primitive][calibrated-probabilities][per-answer-confidence][parallel-questions-per-request][python-and-javascript-sdks]

Technical Specifications

Architecture

ArchitectureNot disclosed. Trained with RLCD (reinforcement learning for calibrated decisions), TypeSafe’s own post-training path, so the model returns decisions and probabilities instead of generated text. The same weights serve every account: no per-customer fine-tuning or LoRA.

Resources

Demo

The Lab

location: San Francisco, California, USA

TypeSafe describes itself as an AI lab building machine-native intelligence infrastructure for automation, designed to make decisions within software. It is led by CEO Diogo Almeida, who co-invented RLHF and InstructGPT, with COO Sasha Sheng (previously Meta FAIR) and CTO Erik Gafni. Its models are trained with RLCD, which the lab defines as reinforcement learning for calibrated decisions, so they return typed answers and probabilities rather than generated text.

more about TypeSafe →

Model ID: jev-1.13.0