Hassan trained a Jev-like classifier for $17
Hassan says he fine-tuned Qwen3.5 4B into a Jev-like classifier for $17 and released the weights, data recipe and tutorial. If it holds up, the classification layer is now something you can clone.
Built with it
Hassan trained tev1-4B, a Jev-like classifier, for $17 and open sourced it
Hassan announced tev1-4B-experimental, a classifier fine-tuned on Qwen3.5 4B, and released the weights, the data recipe and a tutorial for training your own. He says it runs on Together serverless at $0.042 per 1M input tokens and $0 per M output. Asked about hallucinated labels, he said using it with regex, the recommended path, should always hold the schema.
I'm releasing everything: the weights, data recipe, & a full tutorial on how to train your own.
Writing criteria
Sydney Runkle posts four open questions on engineering context for Jev
Sydney Runkle listed the questions she has about context engineering for Jev: what it should and should not do, how to turn a problem into state plus questions, how to present that state optimally, and how to calibrate next steps on the returned probabilities. She asked what else belongs on the list. No answers were posted with it.
how can i calibrate next steps on jev's confidence / probabilities?