XNews

TypeSafe AI Launches Jev: System One Model for Decisions

TypeSafe AI Launches Jev: System One Model for Decisions

TypeSafe AI, founded by ChatGPT co-inventor Diogo Almeida, launched Jev, a System One model that outputs structured decisions with calibrated probabilities instead of text. Jev is 20-200x faster and 40-400x cheaper than frontier LLMs for decision tasks, with 70-500ms response times. It uses RLCD training and is designed for embedding in software as intelligent if statements.

Key insights

Architecture and Speed Advantage

Jev uses parallel sampling instead of autoregressive token generation, enabling all outputs to be computed at once. This results in response times of 70 to 500 milliseconds, compared to 3 to 329 seconds for frontier LLMs. The architecture eliminates hallucination by not generating free text; outputs are type-safe structured results with calibrated probabilities. Type errors are mathematically impossible due to architectural constraints.

RLCD Training vs RLHF

Jev is trained using Reinforcement Learning for Calibrated Decisions (RLCD), a new method developed by TypeSafe. Founder Diogo Almeida argued that RLHF inherently produces hallucination because models learn to discard uncertainty to please the reward model. RLCD avoids this by training for calibrated decisions rather than human-pleasing text. The model outputs an epistemically honest probability with each decision.

Pricing and Cost Efficiency

Input token price is $0.042 per million tokens (about $42 per billion tokens), and output tokens are free. TypeSafe claims Jev is 40 to 400 times cheaper than equivalent intelligence LLMs. The company states it is already profitable and plans to continue lowering prices. The pricing strategy is based on the Jevons paradox: as cost per decision drops, total usage will explode.

Use Cases: Intelligent If Statements

Jev is designed for embedding in software as a fast decision component. Use cases include: intelligent branching in workflows (routing tickets, assessing urgency), large-scale data processing (classification, extraction), real-time applications (game AI, dynamic pricing), and guardrails for LLM outputs (safety checks, format validation). It is not a replacement for conversational models but a complement for structured decisions.

Demos: DOOM and Wikiracing

TypeSafe demonstrated Jev controlling a DOOM AI bot with about 10 queries per second, costing roughly $7 per hour. In a Wikipedia racing task (navigating from one article to another via links), Jev outperformed non-reasoning LLMs in both steps and speed. These demos illustrate Jev's ability to make rapid, repeated decisions in real-time environments.

Philosophical and Market Positioning

The name Jev references the Jevons paradox, reflecting the bet that cheaper AI decisions will massively increase consumption. TypeSafe positions Jev as a new category: System One models (fast, intuitive, automatic) as opposed to System Two (slow, deliberate reasoning). The company's mission is to provide an interface that software can depend on, moving beyond the limitations of RLHF-based models.

Source posts