OVO-S-Bench - Allocentric Spatial Mapping: leaderboard

Metric: Level 4 (allocentric spatial mapping: global direction, topology, trajectory-map alignment) multiple-choice accuracy (%) on the 1,680-question evaluation set, prefix-access mode (128 frames sampled uniformly from the video before the query timestamp, question given with the frames), mean of its three task families. Source: arxiv.org. Saturation forecast: Around 2032. 31 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)54.9
2Qwen 3.5 27B (Non-reasoning)47.72
3Qwen 3 VL 235B A22B45.71
4Qwen 3.5 397B A17B45.36
5Qwen 2.5 VL 7B Instruct44.7
6Gemini 3.1 Flash Lite42.84
7Qwen 3 VL 4B Instruct41.35
8Qwen 3 VL 32B Instruct41.22
9GPT-5.440.53
10InternVL3.5-8B39.32
11Qwen 3.5 4B (Non-reasoning)38.73
12Qwen 3.5 9B (Non-reasoning)36.61
13Grok 4.1 Fast (Non-reasoning)35.04
14Gemma 4 E4B32.26
15Gemma 4 E2B29.62

Interactive version: theaggregate.ai/benchmark?slug=ovo-s-bench-allocentric-spatial-mapping · How It Works · Data refreshed daily, snapshot 2026-09-29.