XNews

TypeSafe AI Launches 'System One' Model Jev for Fast Decisions

TypeSafe AI, founded by ChatGPT co-inventor Diogo Almeida, launched Jev, a new category of AI model called System One. Jev abandons text generation for fast, calibrated decisions in code, with response times of 70-500ms and costs 20-200x cheaper than frontier LLMs.

Key insights

Model Design and Training

Jev is a 'System One' model designed for fast, intuitive decisions, inspired by Daniel Kahneman's thinking. It uses a new training method called RLCD (Reinforcement Learning for Calibrated Decisions). Unlike LLMs, Jev outputs structured results with calibrated probabilities, not text. It cannot generate text, but outputs type-safe structured results with zero type errors by architecture.

Performance and Cost

Jev achieves 70-500ms response time, 20-200x faster than frontier LLMs. Input token price is $0.042 per million tokens, output is free. It uses parallel sampling instead of sequential token generation. The company claims it is 40-400x cheaper than equivalent intelligence models, though extreme speedups are best-case.

Use Cases and Applications

Jev is designed for 'smart if statements' in code: classification, scoring, routing, extraction. Use cases include intelligent branching in customer service, large-scale data processing, real-time applications (game AI, dynamic pricing), and as a guardrail for LLM outputs. It can be called ~10 times per second in a DOOM demo, costing ~$7/hour.

Comparison with LLMs

Jev is not a replacement for conversational models like ChatGPT or Claude. It fills the gap for embedding AI judgment in code where millisecond response and structured output are required. The model's architecture eliminates hallucinations and ensures type safety. It is available in early access via waitlist.

Source posts