EPIC-Bench - Contact Relationship: leaderboard
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 2029. 89 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 53.7 |
| 2 | Gemini 3 Pro | 51.78 |
| 3 | GPT-5.5 (Non-reasoning) | 49.67 |
| 4 | Qwen 3.5 27B | 49.56 |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 48.04 |
| 6 | Qwen 3.6 Plus | 47.84 |
| 7 | Gemini 3 Flash (Preview) | 46.51 |
| 8 | Claude Sonnet 4.6 (Thinking) | 46.21 |
| 9 | Qwen 3.5 122B A10B | 46.04 |
| 10 | Gemini 3 Flash (Preview) (Non-reasoning) | 42.29 |
| 11 | Qwen 3.5 35B A3B | 41.7 |
| 12 | GPT-5 Mini | 41.32 |
| 13 | GPT-5.4 (Non-reasoning) | 39.55 |
| 14 | Qwen 3 VL 30B A3B (Thinking) | 38.78 |
| 15 | Qwen 3.6 27B (Non-reasoning) | 38.69 |
Interactive version: theaggregate.ai/benchmark?slug=epic-bench-contact-relationship · How It Works · Data refreshed daily, snapshot 2026-10-07.