Embodied3DBench - Manipulation Trajectory (2D): leaderboard

Embodied3DBench (2026) Manipulation Trajectory (2D) task on simulated robotic tabletop scenes: end-effector key trajectory points in the image plane, predicted zero-shot by a vision-language model. Normalized 0-100 score derived from box IoU or point distance; higher is better.

Metric: Manipulation Trajectory (2D) task (end-effector key trajectory points in the image plane), 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 Instruct37.6#1332
2GPT-532#178
3Claude Sonnet 426.4#410
4Qwen 2.5 VL 7B Instruct26.3#1902
5GPT-4o24.6#751
6Doubao-1.5-thinking-vision-pro22.6#344
7InternVL3.5-8B22#1604
8Qwen 2.5 VL 72B Instruct21.9#841
9Seed-1.6Non-reasoning20.7#469
10Gemini 2.5 Flash20.1#463
11Gemini 2.5 Pro19.6#276
12Doubao-Seed-1.6Thinking19.3
13InternVL3-38B18.6#939

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

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