The startup TypeSafe AI has unveiled Jev, an artificial intelligence model designed for machine-to-machine communication rather than human interaction. Unlike traditional large language models, Jev generates probabilistic decisions in structured formats like JSON, which are intended for use by other software systems.
The model is based on the Reinforcement Learning for Calibrated Decisions (RLCD) architecture. It is designed for real-time automation, data classification, and acting as a wrapper for other AI models. The developers claim that Jev is not prone to hallucinations, although errors in structured probability outputs remain possible.
Jev demonstrates significant performance advantages, with response times ranging from 70 to 500 ms, making it 40 to 200 times faster than traditional systems. In a demonstration, Jev generated a response in 0.114 seconds, compared to 8.566 seconds for the OpenAI GPT-5.6 Terra model.
The model is also cost-effective, priced at $0.042 per 1 million input tokens, with no charge for output tokens. In comparison, GPT-5.6 Terra costs $2 per 1 million input tokens and $12 per 1 million output tokens. Jev is reportedly 238 times cheaper than the Anthropic Fable 5.1 model.
