EPIC-Bench - Affordance Region: leaderboard
Metric: Manipulation score (0-100) on affordance region localization (the operable part of an object for a described task), 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 November 2027. 89 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 30.4 |
| 2 | Gemini 3.1 Pro (Preview) | 28.89 |
| 3 | GPT-5.5 (Non-reasoning) | 24.62 |
| 4 | Gemini 3 Flash (Preview) | 22.62 |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 16.96 |
| 6 | Qwen 2.5 VL 72B Instruct | 14.77 |
| 7 | Qwen 3.5 Plus | 13.82 |
| 8 | Qwen 3.5 27B | 13.15 |
| 9 | Qwen 3.5 397B A17B | 12.4 |
| 10 | Qwen 3.5 122B A10B | 12.31 |
| 11 | Qwen 3 VL 235B A22B Instruct | 12 |
| 12 | Qwen 3.6 Plus | 11.95 |
| 13 | Qwen 3.5 27B (Non-reasoning) | 11.93 |
| 14 | Gemini 3 Flash (Preview) (Non-reasoning) | 11.91 |
| 15 | Seed 1.8 | 10.94 |
Interactive version: theaggregate.ai/benchmark?slug=epic-bench-affordance-region · How It Works · Data refreshed daily, snapshot 2026-10-07.