RefSpatialBench
Progress Over Time
Interactive timeline showing model performance evolution on RefSpatialBench
State-of-the-art frontier
Open
Proprietary
RefSpatialBench Leaderboard
6 models
| Context | Cost | License | ||||
|---|---|---|---|---|---|---|
| 1 |  Alibaba Cloud / Qwen Team | 28B | 262K | $0.60 / $3.60 | ||
| 2 | Alibaba Cloud / Qwen Team | 236B | — | — | ||
| 3 | Alibaba Cloud / Qwen Team | 122B | — | — | ||
| 4 | Alibaba Cloud / Qwen Team | 27B | 262K | $0.30 / $2.40 | ||
| 5 | Alibaba Cloud / Qwen Team | 35B | — | — | ||
| 6 | Alibaba Cloud / Qwen Team | 35B | — | — |
Notice missing or incorrect data?
What is RefSpatialBench?
RefSpatialBench evaluates spatial reference understanding and grounding.
RefSpatialBench is a image benchmark evaluating models on spatial reasoning, grounding, and vision tasks. LLM Stats tracks 6 models on this benchmark, scored on a 0–100 scale. The current average is 0.7, with the leader at 0.7.
Compare leaders on the best AI for spatial reasoning, best AI for grounding and best AI for vision leaderboards.
Current leaders
Qwen3.6-27B from Alibaba Cloud / Qwen Team currently leads the RefSpatialBench leaderboard with a score of 0.700 across 6 evaluated AI models.
FAQ
Common questions about the RefSpatialBench benchmark and leaderboard.