The missing primitive was a decision, not more words

For two years the industry scaled one thing: generation. A new model, Jev, does the opposite. It cannot write. You give it context and a fixed set of options; it returns one choice, a score or a yes/no, and nothing else. Its makers claim it is far faster and cheaper than a chat model, with no hallucinated prose because there is no prose.

• The reframe is the keeper: much of what we pay a chat model to "reason through" is classification wearing a conversation. Route this ticket, escalate or not, which of five buckets. Name the decision-shaped problems in your stack and you will find dozens.
• Within days an Apache 2.0 clone, Laya, appeared: about 0.4B parameters, weights in hand, self-hosted. The primitive is not just cheap; it can be yours.

• The speed and adoption numbers come from the vendor and from the video; independent benchmarks are thin so far.
• A model that only picks from your options still needs your options to be right. A bad taxonomy decides confidently and wrong.

Source: Για το Jev (Typesafe AI), μοντέλο που μόνο αποφασίζει, και τον ανοιχτό κλώνο του Laya (YouTube)