Skip to content

Benchmark

llminfo Elo (Rapidata votes): Relative Camera Movement Benchmark (Image-to-Video) – Alignment

Scores are Elo ratings. Higher ratings are better.

Computed by llminfo from Rapidata's published CC BY 4.0 vote data. Bradley-Terry ratings on the conventional Elo scale, fitted per board so that the rated models average 1000. Each board has its own scale: a rating is comparable only with others on the same board, not across boards or with other Elo leaderboards. The ± figure is a 95% interval from the fit and understates the true uncertainty, since votes on the same prompt are not independent.

Models measured
13
Scores recorded
13
Sources
1
IQ weight
No weight published

Score distribution

1,027.5
Mean
1,029
Median
48.7
Std dev
969.3
10th percentile
1,098.7
90th percentile
918.9
Min
918.91,099.8
The band covers the middle 80% of scores. The line marks the median.

Leaderboard

Data credits: Ratings computed by llminfo (Bradley-Terry) from human pairwise-vote data published by Rapidata on Hugging Face, CC BY 4.0 — not Rapidata's own ranking. SVG board prompts derive from yupp-ai/yupp-svg-20251204 (Yupp AI, CC BY 4.0); Physics board scenarios are from Physics-IQ (Motamed et al., Google DeepMind / INSAIT). (licence)