AI’s next big leap may not be another chatbot that writes better answers—it could be a model that decides what should happen next.
That is the idea behind Jev, a new AI system from TypeSafe AI that takes a radically different approach to machine intelligence.
Instead of generating paragraphs of text, Jev is designed to make fast, structured decisions, complete with confidence scores, giving AI agents a potential “decision layer” beneath the language models that power them.
Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay
Launched on September 15, Jev is described by TypeSafe AI as a transformer-based model focused on “calibrated decisions”.
ALSO READ : Beyond The AI Hype: Health Tech To Auto Comps — Why Is Rushabh Sheth Pivoting Toward Tactical Bets?
It can help AI agents determine which tool to use next, whether a task should be retried or how a particular input should be classified.
The company says Jev can return these decisions in less than half a second, charging $0.042 per million input tokens, while its output is free.
The model has already been integrated into platforms including Vercel, Cloudflare, LangChain and Langfuse.
TypeSafe AI said on September 21 that Jev was made available without a waitlist, with access beginning at $5 in credits, equivalent to about 120 million tokens.
ALSO READ : China, US Discuss AI, Investment On Second Day Of Trade Talks
How is Jev different from an LLM?
Traditional LLMs are primarily built to understand and generate human language. Jev, by contrast, is designed to operate at the decision-making layer of AI applications.
Developers provide the model with context and define the possible outputs in advance. Jev then selects among those options and assigns probabilities indicating its level of confidence.
TypeSafe AI says the model is focused on “System One” intuition rather than the slower, reasoning-oriented approach used by many advanced AI models.
The company also says Jev does not hallucinate in the conventional sense because it is not generating open-ended text.
Jev was developed by Diogo Almeida, a former OpenAI researcher whose work contributed to instruction-following systems and reinforcement learning from human feedback (RLHF).
Almeida has argued that AI development has largely focused on optimising systems for human language, whereas automation requires machines to make reliable decisions in a more structured environment.
What can Jev be used for?
The model’s potential applications include software automation, classification, model routing and monitoring AI agents.
Developers can use it to screen commands, classify emails, select AI models for particular tasks or assess whether an agent’s action should proceed.
Its low cost and speed could be particularly relevant for companies running large numbers of AI agents, where repeatedly using expensive LLMs can increase operating costs.
Early developer experiments have also reported faster processing and lower costs in some classification tasks compared with larger AI models, although these are individual tests rather than independent benchmarks.
The broader proposition behind Jev is that as AI agents become more widespread, a lightweight decision-making model could work alongside larger language models—handling routine decisions while reserving expensive LLM calls for tasks that require deeper reasoning.
Essential Business Intelligence,
Sharp Market Insights,
Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
