Embodied3DBench - Affordance (2D): leaderboard

Embodied3DBench (2026) Affordance (2D) task on simulated robotic tabletop scenes: functional contact point and interaction vector projected to the image, predicted zero-shot by a vision-language model. Normalized 0-100 score derived from box IoU or point distance; higher is better.

Metric: Affordance (2D) task (functional contact point and interaction vector projected to the image), 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: Not forecast. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-566.1#178
2Qwen 2.5 VL 72B Instruct66.1#841
3Qwen 2.5 VL 7B Instruct55#1902
4Claude Sonnet 454.6#410
5GPT-4o53#751
6Qwen 3 VL 4B Instruct51.9#1332
7Gemini 2.5 Pro50.6#276
8Gemini 2.5 Flash50.3#463
9Doubao-Seed-1.6Thinking44
10Doubao-1.5-thinking-vision-pro41.8#344
11InternVL3-38B41.2#939
12Seed-1.6Non-reasoning38.2#469
13InternVL3.5-8B37.9#1604

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

Interactive version: theaggregate.ai/benchmark?slug=embodied3dbench-affordance-2d · How It Works · Data refreshed daily, snapshot 2026-11-21.