EPIC-Bench - Spatial-Related Attributes: leaderboard
Metric: Target localization score (0-100) on the spatial-attribute tasks (targets described by position, orientation and spatial relations), 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 2030. 89 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 56.65 |
| 2 | Qwen 3 VL 235B A22B (Thinking) | 55.67 |
| 3 | GPT-5.5 (Non-reasoning) | 55.37 |
| 4 | Gemini 3 Pro | 54.5 |
| 5 | Qwen 3.5 Plus | 51.9 |
| 6 | Qwen 3.5 397B A17B | 51.82 |
| 7 | Qwen 3.6 Plus | 50.17 |
| 8 | Qwen 3.5 397B A17B (Non-reasoning) | 50.06 |
| 9 | Qwen 3.5 27B | 50.03 |
| 10 | Qwen 2.5 VL 72B Instruct | 49.79 |
| 11 | Seed 1.8 | 49.68 |
| 12 | Qwen 3 VL 30B A3B (Thinking) | 49.37 |
| 13 | Qwen 3.5 122B A10B (Non-reasoning) | 49.28 |
| 14 | Qwen 3.5 35B A3B | 49.18 |
| 15 | Qwen 3.5 122B A10B | 49.13 |
Interactive version: theaggregate.ai/benchmark?slug=epic-bench-spatial-related-attributes · How It Works · Data refreshed daily, snapshot 2026-10-07.