TypeSafe AI launched Jev, a Transformer model bypassing LLM text generation for software automation. Founder Diogo Almeida argues language optimization suits humans, not machines. Jev outputs probabilities instead of text, eliminating hallucinations and reducing costs. It charges only for input tokens while output tokens are free. Tests show Vercel achieved 5 to 18 times faster speeds with higher accuracy using Jev for command security classification compared to OpenAI. Bryo AI reported costs 10 to 20 times lower than Gemini, with slightly lower accuracy. This approach targets deterministic decision-making for coding tasks rather than creative text generation.