JevBench scores AI models on intelligence, calibration, speed and cost

JevBench targets models that make bounded software decisions rather than produce open-ended prose, according to a September 19 post. The post says the benchmark combines intelligence, calibration, speed and cost using a geometric mean, so exceptional performance in one area cannot fully offset weakness in another. In the comparison described, Jev 1.13.0 leads the composite even though GPT-5.6 Luna has substantially higher hard-case accuracy. The post cautions that this narrow result is not evidence that Jev is generally more capable.