imajev4B Author Benchmark - DecisionBench1.0full23900: leaderboard

imajev4B Author Benchmark - DecisionBench1.0full23900 — Author phase3 current4B adapter and readout, September26 evaluation. Four option rotations plus calibration unless panel explicitly single-pass. Previous checkpoint/comparison references not emitted as another public model. Mac MLX-specific rows retained pending artifact binding, not inherited to PyTorch release. HannoLabs ownharness4rotationscalibrated64ktokens everyrow scored. Officialregistryc9e5f132; distinct from other DecisionBench protocols. Metric: Accuracy (%). Source: https://huggingface.co/mohit67890/imajev-4b

Metric: Accuracy (%). Source: huggingface.co. 1 model tracked. Updated 16 Nov 2026.

Top models

#ModelScoreOverall rank
1imajev-4B79.7

Interactive version: theaggregate.ai/benchmark?slug=imajev4b-author-benchmark-decisionbench1-0full23900 · How It Works · Data refreshed daily, snapshot 2026-11-16.