RefSpatialBench: leaderboard

Reference-based spatial benchmark evaluating whether multimodal models understand relative positions and spatial language in images.

Metric: RefSpatialBench (self-reported). Source: benchmarklist.com. Status: saturation imminent. 4 models tracked.

Top models

#ModelScore
1Qwen 3.6 27B70
2Qwen 3 VL 235B A22B (Thinking)69.9
3Qwen 3.5 122B A10B69.3
4Qwen 3.5 27B67.7
5Qwen 3.6 35B A3B64.3
6Qwen 3.5 35B A3B63.5
7Gemma 4 31B4.7

Interactive version: theaggregate.ai/benchmark?slug=refspatialbench · How It Works · Data refreshed daily, snapshot 2026-09-05.