EPIC-Bench - Affordance Region: 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 the operable part of an object for a described task, on a 0-100 scale; higher is better.

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 May 2027. 76 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro30.4#146
2Gemini 3.1 Pro (Preview)28.89#102
3GPT-5.5Non-reasoning24.62#56
4Gemini 3 Flash (Preview)22.62#156
5InternVL3.5-38B17.22#1026
6Qwen 3 VL 235B A22B (Thinking)16.96#462
7InternVL3.5-241B-A28B16.85#739
8Qwen 2.5 VL 72B Instruct14.77#849
9Qwen 3.5 Plus13.82#250
10Qwen 3.5 27B13.15#380
11Qwen 3.5 397B A17B12.4#270
12Qwen 3.5 122B A10B12.31#350
13Qwen 3 VL 235B A22B Instruct12#541
14Qwen 3.6 Plus11.95#245
15InternVL3.5-14B11.23#1519

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=epic-bench-affordance-region · How It Works · Data refreshed daily, snapshot 2026-11-22.