Jev: the new non-LLM AI model speeding up automation
Alex da Cruz
Alex da Cruz is a full-stack developer based in São Paulo, Brazil. He works with React, TypeScript and automation, and uses AI daily to solve real problems in code and operations — not as a demo. He has run an e-commerce operation end to end, and now builds and maintains the automation pipeline behind this blog. He writes about what he actually tests.
Segundo o TechCrunch, o ex-pesquisador da OpenAI Diogo Almeida lançou o Jev, um modelo da TypeSafe que abandona a geração de texto em favor de decisões probabilísticas.
Why did the creators skip human language?
Large language models optimize for human speech, which makes them inefficient for software automation. Jev outputs calibrated decisions rather than words. Because developers predefine the exact output structure, the model cannot hallucinate text.
How much faster is Jev in practice?
Engineering teams report massive efficiency jumps. Vercel replaced OpenAI's Luna model with Jev for safety classification tasks and achieved results 5 to 18 times faster. Meanwhile, Bryo AI tested Jev against Gemini for email sorting, finding Gemini 10 to 20 times more expensive.
Input tokens on Jev are metered by the billion rather than the million, while output tokens are entirely free. This pricing structure allows developers to use Jev for real-time model routing and monitoring other AI agents without inflating operational budgets.
Sources
Frequently asked questions
- What is Jev?
- Jev is a transformer-based model developed by TypeSafe that outputs probabilities and calibrated decisions instead of generating human text.
- Why does Jev not hallucinate?
- Because the model does not output open-ended text and users predefine the allowable outputs, text hallucination is entirely eliminated.
- How does Jev's pricing work?
- According to TechCrunch, output tokens are completely free, and input tokens are measured by the billion rather than the million.
Comments
0 comments
Be the first to comment.
Continue Lendo
SynthID Watermarking Changes LLM Safety and Tool Execution
New research shows that AI watermarking alters model safety guardrails and tool execution, introducing unexpected behavioral risks for developers.
Claude Docs and Slides: What Changes in the New Interface
Anthropic combines chat, Cowork, Docs, and Slides into a single interface to eliminate tab switching.
Gemini 3.8 Live runs background APIs without pausing speech
Google DeepMind released Gemini 3.8 Live, allowing background API execution and continuous reasoning during active voice interactions.