CHOOSEAI infra
Jev Model Performance on Graph Reasoning
Radoslaw@radoslaw
This project evaluates the performance of a new Jev model from @typesafeai on a synthetic graph-reasoning benchmark, comparing its accuracy, speed, and cost against another model.
I tested @typesafeai new Jev model on a synthetic graph-reasoning benchmark.
Jev: 3 typed classifications → exact solver: 48/48, 274 ms, ~$65/1M.
DeepSeek-V4.1-Flash direct: 48/48, 1.39 s, ~$439/1M.
Interesting early result: same accuracy here, ~5× faster and ~6.7× cheaper. 48 synthetic cases.