EPIC-Bench: 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 is the average score over all samples on a 0-100 scale; higher is better.

Metric: Average score (0-100) over all samples, each sample scored by its task's weighted mix of localization IoU, counting, path and feasibility scores, 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. 76 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro54.81#145
2Gemini 3.1 Pro (Preview)54.72#99
3Qwen 3 VL 235B A22B (Thinking)50.93#459
4GPT-5.5Non-reasoning50.16#54
5Qwen 3.5 397B A17B47.47#264
6Qwen 3.5 122B A10B47.24#346
7Qwen 3.5 Plus47.1#242
8Doubao-Seed-1.6-vision46.8#361
9Seed 1.846.32#252
10Qwen 3.5 27B45.67#375
11Qwen 3.5 35B A3B45.58#492
12Qwen 3.6 Plus45.48#238
13Qwen 3.6 27BNon-reasoning45.38#351
14Gemini 3 Flash (Preview)44.69#152
15Qwen 3 VL 30B A3B (Thinking)44.02#719

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