Skip to content

Models Benchmarks Providers

Search models and benchmarks /

RefSpatialBench

vision

Categories: grounding, spatial reasoning, vision
Modality: image
Language: en
Multilingual: No
Max score: 100
Scoring: %, higher is better
Verified by llm-stats: No

RefSpatialBench evaluates spatial reference understanding and grounding.

Leaderboard

Showing 6 of 6 results

Qwen3.6-27B

70.0%

i
Qwen3 VL 235B A22B Thinking

69.9%

i
Qwen3.5-122B-A10B

69.3%

i
Qwen3.5-27B

67.7%

i
Qwen3.6-35B-A3B

64.3%

i
Qwen3.5-35B-A3B

63.5%

i

Wikibench About Theme Content licensed CC BY-SA 4.0.