My feed has been overrun by mentions of Jev from TypeSafe AI - Introducing System One Models and Jev and many folks are stuffing it into every use case imaginable while others are claiming that it is “just” another classifier. I did a little bit of experimentation over the weekend:
What it IS:
Jev is closest to what is called a Zero Shot Classifier that is hosted with a low latency API around it (having a classical ML background helps here).
This is not trivial - Classifiers have been around for a long time but a
general purpose Classifier => Any classifier task
zero shot => No need to provide examples first
hosted low latency=> Fast API/SDK Call
ultra low cost=> Much cheaper than most LLMs
unlocks a lot of use cases that AI people find mind-blowing.
What it is NOT:
Typesafe -makers of Jev- brands it as a Type 1 Model - borrowing from the “Thinking, Fast and Slow” book by Daniel Kahneman. This is a bit of stretch - the Model does not think and returns a very narrow set of responses compared to what Kahneman talks about in the Type 1 thinking in his book which is about fast instinctive responses.
The Jev system does not hallucinate like a traditional LLM (it will not for example invent a new classification if you ask it to label things as a Cat or Dog - you will always get the probabilities that something is a Cat or Dog and not say Cow).
However it is still a Classifier trained on data, so:
- It cannot generate text or chat or images (not a generative model)
- It will get probabilities wrong sometimes (it is not a perfect truth-teller)
- It will get its binary classification wrong (confidently so -sometimes)
- It is sensitive to the order of inputs (same inputs but different order might change answers)
- It might give you different probabilities even if you ask the exact same question twice
When to use it:
4 components together make for a good use case
- Natural Language is involved either in the input or for output placement
- LLMs like GPT-5x or Fable can do the task but are too slow/expensive/overkill
- There is limited set of options to choose and you need Jev to make fast decisions between them
- Text generation is not required
How to decide to use it:
Now we step into the land of evals or Evaluations - basically what good looks like in terms of quantifiable metrics. Give the following terms to your agent: Accuracy, Confusion Matrix, F1 Score, Classifiers. Or look at the writings of Hamel Husain or Shreya Shankar
Future directions:
One of more of the big labs will probably introduce something like this and there are already a ton of OSS resources for Classifiers in the form of ModernBERT GliNER, etc. so expect multiple clones and open-source DiY offerings.
BUT Jev’s marketing has been outstanding - so there is a real PFM here.