Jev is a very good classifier wearing a frontier model's coat
Two years in stealth. Ex-OpenAI. Co-invented ChatGPT. That is the pitch for Jev, and the pitch is doing most of the lifting. Strip the marketing off and look at what it actually does. Hand it a list of options, it picks one. Hand it criteria, it scores against them. Hand it a proposition, it rules true or false. That is the whole surface. A narrow-interface JSON classifier. Useful, genuinely. Not…
Jev, a classifier utilizing a frontier model's coat, has been in stealth for two years and was co-invented by an ex-OpenAI employee. This JSON-based classifier excels at picking one option from a list, scoring against criteria, and determining if a proposition is true or false. The "no hallucinations" claim is merely a naming decision, as the model is masked from emitting values outside of a predefined set.
The technology is not novel and can be found in various approaches such as OpenAI Structured Outputs schema-constrained decoding, Anthropic tool use typed input schema, LangChain + Pydantic parse and validate, and retry on failure. It acts as a sidecar to the big LLM, handling routing, moderation, risk scoring, and validation while the large model deals with the messy human request.
The accuracy trade-off is real, as the classifier lags behind a current frontier model in multiple-choice picking. The context window is limited to 32k, which is insufficient for handling complex workflows. Within two hours of its launch, a developer named Harsha Gundala replicated Jev's functionality using Qwen-2.5-1B-RLCD on Hugging Face, demonstrating that the recipe is not a secret.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
