EPIC-Bench - Ground Detection: 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 traversable ground regions, on a 0-100 scale; higher is better.

Metric: Navigation score (0-100) on ground detection (localizing traversable ground regions), 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 March 2027. 76 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3.5 397B A17BNon-reasoning51.94#264
2Qwen 3.5 122B A10B51.42#346
3Gemini 3 Flash (Preview)51.11#152
4GPT-5.5Non-reasoning50.97#54
5Qwen 3 VL 235B A22B Instruct50.96#532
6Qwen 3.5 Plus50.85#242
7Qwen 3.6 27BNon-reasoning49.99#351
8Qwen 3.5 27BNon-reasoning49.74#375
9Qwen 2.5 VL 72B Instruct49.45#841
10Qwen 3 VL 30B A3B Instruct49.37#895
11Gemini 3 Pro48.82#145
12Qwen 3 VL 235B A22B (Thinking)48.18#459
13Qwen 3.5 35B A3BNon-reasoning48#492
14Gemini 3.1 Pro (Preview)47.22#99
15InternVL3.5-14B46.11#1504

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-ground-detection · How It Works · Data refreshed daily, snapshot 2026-11-21.