Embodied3DBench - Correspondence (2D): leaderboard

Embodied3DBench (2026) Correspondence (2D) task on simulated robotic tabletop scenes: cross-view correspondence of a given point in image coordinates, predicted zero-shot by a vision-language model. Normalized 0-100 score derived from box IoU or point distance; higher is better.

Metric: Correspondence (2D) task (cross-view correspondence of a given point in image coordinates), 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
1Qwen 3 VL 4B Instruct58.3#1332
2Qwen 2.5 VL 72B Instruct55.7#841
3GPT-553.9#178
4Claude Sonnet 437.5#410
5Qwen 2.5 VL 7B Instruct36.2#1902
6Gemini 2.5 Flash32.2#463
7InternVL3.5-8B31#1604
8Gemini 2.5 Pro30.7#276
9GPT-4o30.5#751
10Doubao-1.5-thinking-vision-pro28.4#344
11InternVL3-38B27.9#939
12Seed-1.6Non-reasoning27.3#469
13Doubao-Seed-1.6Thinking26.7

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

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