EPIC-Bench - Contact Relationship: leaderboard

EPIC-Bench (2026) is a mask-grounded benchmark of embodied visual perception: 6,661 annotated image, text and mask samples across 23 tasks in target localization, navigation and manipulation, scored by localization, counting, path and feasibility metrics. This board scores localizing and counting the objects in contact with a reference object, on a 0-100 scale; higher is better.

Metric: Manipulation score (0-100) on contact relationship (localizing and counting objects in contact with reference objects), on EPIC-Bench (6,661 human-annotated image, text and mask tuples from 25 public datasets across 23 fine-grained embodied perception tasks; the model outputs bounding boxes, counts, path points or feasibility judgements that are scored against the masks), zero-shot, averaged over six runs for local models and two for API models; higher is better. Source: arxiv.org. Saturation forecast: Around June 2028. 76 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.1 Pro (Preview)53.7#99
2Gemini 3 Pro51.78#145
3GPT-5.5Non-reasoning49.67#54
4Qwen 3.5 27B49.56#375
5Qwen 3 VL 235B A22B (Thinking)48.04#459
6Qwen 3.6 Plus47.84#238
7Doubao-Seed-1.6-vision47.48#361
8Gemini 3 Flash (Preview)46.51#152
9Claude Sonnet 4.6Thinking46.21#173
10Qwen 3.5 122B A10B46.04#346
11Qwen 3.5 35B A3B41.7#492
12GPT-5 Mini41.32#364
13GPT-5.4Non-reasoning39.55#135
14Hunyuan-T1-Vision38.95#767
15Qwen 3 VL 30B A3B (Thinking)38.78#719

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

Interactive version: theaggregate.ai/benchmark?slug=epic-bench-contact-relationship · How It Works · Data refreshed daily, snapshot 2026-11-21.