IDEAL-Bench - Perceptual Score: leaderboard
Render-and-compare evaluation of IDEAL-Bench (2026): each predicted 3D layout is re-rendered and a Claude Opus 4.7 judge rates its spatial similarity to the ground-truth image from 1 to 5, over 200 scenes (20 per room type) and five judging passes. Higher is better.
Metric: Mean perceptual score (1-5): a Claude Opus 4.7 judge compares each scene re-rendered from the predicted layout with the ground-truth image and rates spatial similarity on a 1 to 5 Likert scale, 200 scenes (20 per room type), mean of five judging passes; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 3.56 | #266 |
| 2 | GPT-5.4 | 3.49 | #123 |
| 3 | Gemma 4 31B (IT) | 3.22 | #342 |
| 4 | Claude Sonnet 4.6 | 3.06 | #161 |
| 5 | GPT-4o | 2.78 | #741 |
| 6 | Qwen 2.5 VL 72B Instruct | 1.92 | #846 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ideal-bench-perceptual-score · How It Works · Data refreshed daily, snapshot 2026-11-13.