Spatial-SSRL-Qwen3VL-4B: benchmark results
Provider: Shanghai AI Lab. Released 2025-11-24. Access: Open.
Aggregate Rating 1634 ± 45, rank #732 of 10916 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MJev-Spatial - RoboSpatial Pointing | 61.48 | Any-first-two point mask success (%) | 100 |
| Spatial-SSRL4B Release - General VQA | 70.28 | Average score (%) | 100 |
| Spatial-SSRL4B Release - Spatial understanding | 61.43 | Average score (%) | 100 |
| SpatialQuery-1M - Closest-Instance Distance | 42 | Acc@0.2m (%; share of 5,000 fixed closest-instance distance | 100 |
| SpatialQuery-1M - Proximity Decision | 76.04 | Accuracy (%; binary decision whether the closest same-catego | 100 |
| AeroGround - Reverse Sequence Reasoning | 43.8 | Accuracy (%; Spatial Reasoning: find the ground frame that d | 86.7 |
| MJev-Spatial - EmbSpatial | 78.1 | Accuracy (%) | 71.4 |
| MJev-Spatial - Overall | 60.08 | Twelve-benchmark arithmetic mean (%) | 71.4 |
| MJev-Spatial - Where2Place | 66.54 | Mean grayscale mask score (%) | 71.4 |
| AeroGround - Multi-Target Relation | 60.6 | Accuracy (%; Spatial Understanding: order comparison targets | 66.7 |
| AeroGround - Relative Orientation Determination | 28.9 | Accuracy (%; Spatial Understanding: select the egocentric qu | 63.3 |
| AeroGround - Single-Target Matching | 35.4 | Accuracy (%; Object Alignment: select the candidate in the o | 60 |
Interactive version: theaggregate.ai/model?slug=spatial-ssrl-qwen3vl-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.