Last week a company called TypeSafe AI released a new model called Jev
It cannot write an essay and it will not chat. Ask it something and it returns a decision: yes or no, which category, a number, together with a probability of how sure it is.
That sounds like a step backwards. But the company claims that, for the same decision, it is 40 to 200 times faster and 40 to 400 times cheaper than today's frontier LLMs: about US$0.0004 a case, answered in 0.4 seconds.
What interests me is not another model. It is the division of labour.
In the AI agents I build, most of the time goes on small judgements rather than writing. Does this email need action now? Which discipline does this drawing belong to? Is this figure over the limit? Asking an LLM to think and write a paragraph for each one is slow and expensive.
Hand those repeated judgements to a model like Jev, and the LLM is left for the work that needs reasoning, writing and explaining. The saving is not a few per cent. It is a few hundred times.
Asking an LLM to think and write a paragraph for each one is slow and expensive.
The name comes from the 19th-century economist Jevons, who noticed that when something becomes cheaper, people use more of it, not less. When AI decisions cost this little, a lot of small jobs we thought "not worth AI" will end up with it.
To be clear: those figures are the company's own, Jev is still in early access, and I have not tried it. But I believe the direction. Bigger is not always better; the right model in the right place is what counts.
In your work, is there a decision you make dozens of times a day that is really just yes or no? Would an AI that only decides, and never talks, change how you work? Tell me what you think in the comments 👇
