NimCritic VPB Training Data - Forward PRC: leaderboard

NimCritic v1 visual manipulation-progress evaluation, not action execution. LOOM: 300 trajectories, 19,618 calibrated frames, 171 single-task and 129 multi-task; duration-weighted Spearman forward/reverse correlations, signed harmonic PRC-F1 (zero on opposite signs) and terminal completion accuracy with reference progress > 0.99. Native baseline interfaces/output definitions retained. GVL shuffles up to 30 frames, temperature 1; GPT-6 Sol extra reasoning disabled. NimCritic is a separately trained Wan2.2-TI2V-5B-initialized critic with frozen UMT5-XXL,128,000 updates and 20 Euler inference steps; Gemini 3.8 Flash decomposes subtasks. Other checkpoint revisions/efforts unreported; no aliases to untrained backbones. Training Data ablation; every row is a separately trained critic, exact three-column arm preserved in model identity. Metric: Forward PRC.

Metric: Forward PRC. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 4 models tracked.

Top models

#ModelScoreOverall rank
1NimCritic Training Data [✓ \checkmark, ✓ \checkmark, ✓ \checkmark]0.73
2NimCritic Training Data [× \times, × \times, ✓ \checkmark]0.72
3NimCritic Training Data [× \times, ✓ \checkmark, × \times]0.66
4NimCritic Training Data [× \times, × \times, × \times]0.62

Interactive version: theaggregate.ai/benchmark?slug=nimcritic-vpb-training-data-forward-prc · How It Works · Data refreshed daily, snapshot 2026-10-27.