Jev introduces a new shape of LLM - System One, aka Decision Models
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Jev outputs typed probabilistic decisions (category scores, yes/no, ratings with confidence) instead of tokens, charges only for input at $0.042/M tokens with free output, and evaluates many parallel questions against one document in roughly single-question latency. This makes it dramatically cheaper and faster than a generative LLM for classification, prioritization, and reranking (e.g. scoring 100 BM25 candidates for relevance), but you get zero explanation—just a float—so build heavy evals and bias testing into your pipeline and never use it for anything like ranking job applicants.
Jev introduces a new LLM variant that returns floating-point numbers instead of text, enabling classification tasks like spam detection or relevance scoring with high speed and low cost at $0.042 per million input tokens (cheaper than GPT-5 Nano). This makes it ideal for scalable, parallelizable decision-making workflows, though its black-box nature and lack of explainability require rigorous evaluation to mitigate potential biases in production use.