NanoJev
A 0.6B decision model that plays ViZDoom, mazes and Snake without decoding
Emerged finds new products worth a look, daily — what’s new today
Teams paying per call to route tickets or classify text can run the same kind of probability-returning decision model on their own GPU or Mac.
Kev is a family of 0.8B, 4B and 9B models fine-tuned from Qwen3.5 that answer several typed questions about one input at once and return probabilities rather than prose. Its local server speaks the same API as TypeSafe's System One, so the TypeSafe Python SDK can point at it unchanged. It runs on CUDA, ROCm and Apple Silicon, ships training code, frozen evaluation suites, a research log and a web playground, and there is a no-install demo on Hugging Face.
More like this
Bespoke Nimble
An open 9B model, dataset and training recipe for fast typed decisions over text
Laya
Small encoder models that answer routing and triage questions in one pass
SemIf
Read a yes/no probability straight out of a small open model, no generated text
jev-align
Teach a Jev classifier your labels by reviewing only its least certain answers
Flide
Exact, shared-context inference server for small Jev-style decision models