Live demo
Watch Jev play Wikiracing
Start on one Wikipedia article, reach another using only the links you meet. Every hop is a single Choice question over the links on the current page — which makes this the clearest thing you can watch a decision model do.
Why a game is the honest demo
A game is a long run of decisions over an action space that changes every step. That is the exact shape Jev exists for, which is why TypeSafe reached for games in their own launch post rather than a benchmark table.
It is also unusually hard to fake. You can see every option it was offered, the probability it gave each one, and what the move cost — all in the same view.
TypeSafe's own two demos
Wikiracing
Start on one Wikipedia article and reach a target article using only the links you meet along the way. Every hop is one Choice question over the links on the current page — and "each step can mean choosing between hundreds to thousands of links".
Jev "tended to finish in fewer steps" than the LLM baselines.
The speed advantage was smaller here than in their other demos, because this compared against non-reasoning LLM modes. And it runs straight into the 255-option ceiling, which is why a two-stage funnel is needed.
Doom
The bot reads a structured game state as a data structure of text, not pixels — TypeSafe notes "not on images (yet…)" — and picks its next action from a fixed set.
Roughly 10 queries per second, which they put at about $7/hour.
TypeSafe say plainly that "a non-AI doom bot could play better". The demo is about instruction-following at speed, not about winning.
Doom is the more spectacular of the two but the harder one to host: at roughly ten queries a second it runs about $7/hour, which is why the demo on this page is Wikiracing. Note also that Doom reads a text state, not pixels — text only — no image or audio input at launch
The 255-option ceiling
A Wikipedia article can link to hundreds or thousands of others, and Choice accepts at most 255 options. TypeSafe handle this with a two-stage system. This page takes the cheaper route and shortlists 48 links per hop — the options are the prompt, so every extra link is input tokens on every move.
Which is the real lesson for anyone building on a decisions model: the candidate set is yours, and it is where your bugs will live. Jev can only be as good as the options you hand it.
Article data from the Wikipedia API. Demos referenced from TypeSafe's launch post.