Jev
Jev is a System One Model from TypeSafe AI designed for fast, structured decision-making inside software and automated workflows. Unlike conventional large language models that generate free-form text token by token, Jev accepts unstructured state and returns predefined type-safe values with calibrated probabilities and confidence scores. The model uses a parallel sampling architecture and a training approach called Reinforcement Learning for Calibrated Decisions, which is optimized for producing reliable probabilistic decisions rather than conversational responses. Jev is intended for tasks such as classification, routing, scoring, extraction, branching logic, verification, guardrails, and other AI-powered workflows where structured outputs can plug directly into application code. TypeSafe reports end-to-end response times ranging from about 70 to 500 milliseconds and says Jev can be substantially faster and more efficient than frontier language models on System One-style workloads.
Learn more
Clef
Clef is a 27B multimodal decision model that turns a state and a schema of typed questions into structured decisions. It can read state information as text, JSON, images, or video and return a probability for every allowed option of every question. The model supports a 65,536-token context window and vision inputs, allowing developers to evaluate both structured application data and visual content in the same decision workflow. Clef supports three typed question formats: noul for estimating the probability that a proposition is true, choice for selecting among predefined options and returning per-option probabilities, and score for evaluating an input against ordered criteria and producing a probability-weighted score. Applications can submit between 1 and 64 questions in a single request, with answers returned under the same question IDs.
Learn more
Laya AI
Laya AI is an open source family of non-autoregressive System 1 decision models built for fast, typed decisions instead of free-form text generation. Given a state and structured questions, Laya evaluates them in a single forward pass and returns outputs that application code can use directly. It supports three decision types: choice, which selects among named options and returns probabilities; score, which places an input on an ordered rubric such as urgency or severity; and noul, which estimates the probability that a proposition is true. Laya is designed for workloads such as agent and model routing, support ticket triage, guardrails, intent classification, risk and fraud scoring, workflow escalation, RAG filtering, moderation, and phishing detection. It can route requests by department, score urgency, flag refund or churn signals, detect policy or abuse conditions, and escalate uncertain cases to a larger model or human reviewer.
Learn more
Liquid d1
d1 is Liquid AI’s first decision model, built to answer structured questions about a situation with probabilities rather than generating text. It takes unstructured data such as text, images, or both, along with one or more questions, processes them in a single forward pass, and returns probabilities without generating output tokens. d1 supports three question types: noul for yes-or-no decisions, choice for selecting among labels with a probability for each option, and score for placing an input on an ordered scale using probability-weighted levels. Multiple questions can be evaluated against the same state in one request, making it useful for real-time decision workflows. Liquid AI demonstrates d1 across applications such as filtering support tickets, browsing code repositories, organizing documents into smart folders, operating websites, compacting agent context, inspecting industrial parts, and playing games from screenshots.
Learn more
