spatial reasoning benchmark
RefSpatialBench
RefSpatialBench evaluates spatial reference understanding and grounding.
Updated Aug 7, 2026
Published models6
Registry coverage6
MetricScore
EvidenceB
RefSpatialBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 0.7 points | 100.0% | 6 | C | |
| 2 | AC | 0.699 points | 80.0% | 6 | C | |
| 3 | AC | 0.693 points | 60.0% | 6 | C | |
| 4 | AC | 0.677 points | 40.0% | 6 | C | |
| 5 | AC | 0.643 points | 20.0% | 6 | C | |
| 6 | AC | 0.635 points | 0.0% | 6 | C |
Score distribution
Top published rows on the benchmark's original scale.
RefSpatialBench highlights
The top published results on this benchmark's own scale.
What is RefSpatialBench?
Definition and scoring fields from the benchmark registry.
RefSpatialBench evaluates spatial reference understanding and grounding.
Scores are shown in points. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- RefSpatialBench
- Modality
- image
- Primary category
- spatial reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- refspatialbench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about RefSpatialBench.
Which model scores highest on RefSpatialBench?
Qwen3.6-27B is currently ranked first with 0.7 points.
What does RefSpatialBench measure?
RefSpatialBench evaluates spatial reference understanding and grounding.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
6 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.