Laya - the OS version of Jev, created a year ago

AI Forum AI In The News
, , ,
1

Evaluates typed decisions (choice, score, noul) over 100+ languages in a single forward pass with calibrated probabilities. Outperforms TypeSafe Jev.

And then in September 2026, a well-funded frontier lab called TypeSafe AI (founded by Diogo Almeida, a co-inventor of ChatGPT at OpenAI) launched Jev. They proposed the exact same non-autoregressive decision concept as if it was a brand-new scientific breakthrough. Except they launched without technical papers, without open weights, and with zero open training datasets.

Read in full here:

1 Like
2

Whether you evaluate Jev or Laya, the decision-model economics still reward keeping serialized state lean. Jev’s roughly $0.042/M input-token pricing with output free makes filler turns and duplicate tool observations a recurring cost.

A practical Tier-1 step is local deterministic compression before the state crosses the wire. contextpress (pip install contextpress) is worth a look: same input gives the same output, and its agent profile keeps tool call/result pairs intact. I’d start low or medium, then measure decision quality and input tokens per decision.

3

Classification is not a new concept.
Jev remind us using LLM may be be always be the cost effective option.

There are lots of open source ‘Jev’ clones appear in the weekend.
I think we will see even more ‘clones’ in coming weeks.