Jev 1.13.0
AI model · Multimodal model · Text → Text by TypeSafe AI
High-speed System One model for structured text and image analysis.
Jev 1.13.0 is a stable version of TypeSafe AI's System One model family, optimized for high-speed automated decision-making. It processes text and image inputs to return structured data, including choices, scores, and truth values, which can be integrated directly into software logic. Unlike newer versions, Jev 1.13.0 focuses on Choice, Score, and Noul question types, providing the same calibrated confidence and parallel sampling benefits. It is designed for businesses that require reliable, machine-native intelligence for workflows where traditional chat-based LLMs are too slow or inconsisten
- Languages
- English
- Model ID
- jev-1.13.0
- Per-request Context Length
- 64k tokens
- Per-question Context Length
- 32k tokens
- Input Modalities
- Text, Images
- Question Types
- Choice, Score, Noul
Overview
Jev 1.13.0 is a stable version of TypeSafe AI's System One model family, optimized for high-speed automated decision-making. It processes text and image inputs to return structured data, including choices, scores, and truth values, which can be integrated directly into software logic. Unlike newer versions, Jev 1.13.0 focuses on Choice, Score, and Noul question types, providing the same calibrated confidence and parallel sampling benefits. It is designed for businesses that require reliable, machine-native intelligence for workflows where traditional chat-based LLMs are too slow or inconsistent. The model is accessed via the TypeSafe API and maintains a competitive pricing structure with free output tokens.
AI Primitives
- Choice Question: Select an option from a provided list with associated probabilities.
- Score Question: Evaluate state against a rubric to provide a numerical score.
- Noul Question: Assess the truth of a statement as a probability between 0 and 1.
Core Features
- RLCD Training: Uses Reinforcement Learning for Calibrated Decisions (RLCD) for epistemically honest probabilities.
- Parallel Sampling: Processes all questions against the state in a single parallel query rather than sequentially.
- Confidence Reporting: Provides calibrated confidence estimates to help software decide when to escalate to human review.
Pricing
- Input tokens (per billion)
- $42
- Input tokens (per million)
- $0.042
- Output tokens
- Free
Specifications
- Model ID
- jev-1.13.0
- Per-request Context Length
- 64k tokens
- Per-question Context Length
- 32k tokens
- Input Modalities
- Text, Images
- Question Types
- Choice, Score, Noul
Added by AntSpace from public sources (docs.typesafe.ai).