Jev model explosion and the hunger for smaller decision models

Multiple threads today circled the same new thing: Jev-like decision models, small models trained specifically for classification and routing rather than generation. There was Kev, a Jev-like family built on Qwen3.5, a thread on turning Jev into a chatbot, and Jev-Leftpad, which someone built seemingly just to see if they could. Commenters noted 'a bit of a Jev explosion going on' and asked whether this is nostalgia for simpler, more understandable systems, noting 'classification models have been around for a while.'

The pattern here is real and worth watching. As agentic pipelines get more complex, people are looking for lightweight models that can make routing decisions cheaply and predictably, without burning through tokens on a full reasoning model. One commenter said Kev 'could really simplify some of our internal routing logic.' The interest isn't in Jev specifically, it's in the broader question of whether you need a sledgehammer when a scalpel will do.

The Show HN for lossless-memory, a personal AI memory tool that never summarizes, fits the same thread: builders are actively working around the failure modes of big general models by building specialized, composable pieces instead.