Former OpenAI Researcher Introduces Jev, a Non-LLM Transformer Model for Software Automation

Diogo Almeida, a former OpenAI researcher who helped develop ChatGPT and co-invent Reinforcement Learning from Human Feedback (RLHF), has introduced a new transformer-based architecture aimed at software automation. Having left OpenAI two years ago to launch TypeSafe AI, Almeida sought to address the mismatch between human-language generation and machine execution. This week, the startup unveiled Jev, a transformer model that is not a large language model (LLM). Instead of outputting text, Jev produces probabilities and calibrated decisions against options predetermined by developers. By avoiding generative text, Jev eliminates hallucinations entirely, as users define possible outputs in advance. The architecture is significantly faster and cheaper to operate than conventional LLMs, offering free output tokens and metering input tokens by the billion rather than the million. Developer demand for the model briefly interrupted TypeSafe AI's API services. Early enterprise adopters report noticeable efficiency gains; Pranit Sharma, a software engineer at agentic infrastructure platform Vercel, noted that replacing OpenAI's Luna 5.6 with Jev for command safety classification yielded results five to 18 times faster alongside higher accuracy. Additionally, Bryo AI CTO Nikhil Mudholkar tested Jev against Google's Gemini for business email classification.