X · by laxman
Post-train your own judge on your agent's traces
A 4B, 8B or 27B judge that beats Jev on human agreement.
laxman
@llmluthor
Releasing Your Own Jev Post-train a 4B/8B/27B judge on your agent's traces. It beats Jev. 79.7% agreement with human labels vs Jev's 66.3% Beats DeepSeek-V4.1-Flash (763B) by 14 points 0.13s per step on a single GPU Completely open source: recipe, data, training, evals
Sep 30, 2026 · 566 likesOpen on X
laxman post-trains a judge on an agent's own traces and open-sources the recipe, data, training and evals. He reports 79.7% agreement with human labels against Jev's 66.3%, a 14 point lead over DeepSeek-V4.1-Flash (763B), and 0.13s per step on a single GPU.
Open the source
