SpatialEdit-Bench - Camera Overall Error: leaderboard
SpatialEdit-Bench (2026) tests instruction-based spatial image editing: moving, scaling or rotating one object, or changing the camera viewpoint of a scene. This board is the camera overall error, the mean of a 3D pose-based viewpoint error and a detector-based framing error for yaw, pitch and zoom edits, where 0 is perfect and lower is better; video world models are scored on their final frame.
Metric: Camera overall error: mean of the VGGT viewpoint error (pose error, scale-normalized) and the detector framing error, for yaw, pitch and zoom camera edits; 0 is perfect; lower is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | LingBot-World | 0.699 | |
| 2 | ReCamMaster | 0.738 | |
| 3 | LongCat-Image-Edit | 0.743 | |
| 4 | Seedream 4.0 | 0.77 | |
| 5 | Nano Banana | 0.777 | |
| 6 | Qwen-Image-Edit | 0.807 | |
| 7 | Edit-R1 (SpatialEdit-Bench checkpoint unspecified) | 0.824 | |
| 8 | Kling 2.5 Turbo | 0.892 | |
| 9 | Vidu Q2 Turbo | 0.897 | |
| 10 | Veo 3.1 | 1.05 |
Interactive version: theaggregate.ai/benchmark?slug=spatialedit-bench-camera-overall-error · How It Works · Data refreshed daily, snapshot 2026-11-21.