SenseNova-SI (Spatial Intelligence)
model paper datasetOpen-source spatial intelligence model family built on multimodal foundations (Qwen3-VL, InternVL3). Curated SenseNova-SI-8M dataset of 8 million spatial samples. SenseNova-SI-8B outperforms GPT-5 and Gemini-3-Pro on spatial benchmarks including VSI-Bench (68.7%), MMSI (43.3%), MindCube (85.6%). Accepted at CVPR 2026.
Outputs 3
SenseNova-SI Models
modelVariants
| Name | Parameters | Notes |
|---|---|---|
| SenseNova-SI-1.1-Qwen2.5-VL-3B | 3B | — |
| SenseNova-SI-1.1-Qwen2.5-VL-7B | 7B | — |
| SenseNova-SI-1.1-InternVL3-2B | 2B | — |
| SenseNova-SI-1.1-InternVL3-8B | 8B | — |
| SenseNova-SI-1.2-InternVL3-8B | 8B | State-of-the-art among open-source spatial models |
Scaling Spatial Intelligence with Multimodal Foundation Models
paperarXiv: 2511.13719
Venue: CVPR 2026
SenseNova-SI-8M Dataset
dataset8 million diverse spatial data samples under a rigorous taxonomy of spatial capabilities for training spatial intelligence models.