Jev Your Engines: Seven Open Decision Models vs. a Generative LLM

For simple decision tasks like routing support tickets, specialized "Jev-style" models can directly score a list of options, avoiding the need to generate and parse text answers. A hands-on benchmark tested seven of these open-source decision models against a generative LLM to see which could more accurately and efficiently route banking customer requests. The leading decision models, Kev and Verdict, significantly outperformed the generative LLM baseline in routing accuracy, especially when handling dynamically changing choices for each request. These specialized models also returned decisions with dramatically lower latency and generated zero answer tokens, highlighting a major efficiency gain over the LLM's text-based approach.