Embodied3DBench - Spatial Relation (3D): leaderboard

Embodied3DBench (2026) Spatial Relation (3D) task on simulated robotic tabletop scenes: object-to-object and object-to-camera direction and distance in 3D, predicted zero-shot by a vision-language model. Normalized 0-100 score derived from box IoU or point distance; higher is better.

Metric: Spatial Relation (3D) task (object-to-object and object-to-camera direction and distance in 3D), normalized score (0-100): IoU of predicted boxes or the L2 error of predicted points (pixels in 2D, metres in 3D) mapped to 0-100, on the simulated robotic tabletop scenes of Embodied3DBench (21K QA pairs over 4.8K object instances), zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-577.7#178
2Doubao-Seed-1.6Thinking72.2
3Doubao-1.5-thinking-vision-pro67.8#344
4Gemini 2.5 Flash67.4#463
5InternVL3-38B67.3#939
6Claude Sonnet 465.2#410
7Qwen 3 VL 4B Instruct65.2#1332
8Seed-1.6Non-reasoning63.7#469
9GPT-4o63.3#751
10InternVL3.5-8B60.4#1604
11Gemini 2.5 Pro60.2#276
12Qwen 2.5 VL 7B Instruct59.8#1902
13Qwen 2.5 VL 72B Instruct56.5#841

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=embodied3dbench-spatial-relation-3d · How It Works · Data refreshed daily, snapshot 2026-11-21.