EPIC-Bench - Feasible Path: 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 planning a point path to the target that stays on traversable ground, on a 0-100 scale; higher is better.

Metric: Navigation path score (0-100) on feasible path planning (a point path from the camera or a marked region to the target that stays on traversable ground), 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 February 2027. 76 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro70.82#145
2Gemini 3.1 Pro (Preview)70.32#99
3Claude Sonnet 4.6Thinking68.46#173
4Qwen 3.5 122B A10B63.47#346
5Qwen 3.6 Plus62.12#238
6Qwen 3.5 35B A3B61.17#492
7Qwen 3.5 397B A17B60.11#264
8Qwen 3.6 27BNon-reasoning59.15#351
9Doubao-Seed-1.6-vision58.74#361
10Qwen 3.5 Plus57.89#242
11Qwen 3.5 27BNon-reasoning53.5#375
12Seed 1.853.43#252
13Gemini 3 Flash (Preview)52.63#152
14Gemini 2.5 FlashNon-reasoning51.98#463
15Qwen 3 VL 235B A22B (Thinking)51.19#459

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