SenseNova-SI - SenseTime's open-source spatial intelligence model

What is SenseNova-SI?

SenseNova-SI is an open-source spatial intelligence model from SenseTime, focused on enhancing spatial intelligence. Trained on large-scale, high-quality spatial data, the model significantly improves its capabilities in core dimensions such as spatial measurement, relationship understanding, and perspective transformation. In multiple authoritative benchmark tests, SenseNova-SI surpasses other open-source models of similar scale and outperforms top closed-source models like GPT-5. The model provides detailed installation and usage guides to help developers get started quickly, driving the development of embodied intelligence and world models, and laying the foundation for AI to understand the three-dimensional world.

Main functions of SenseNova-SI

• Space Measurement and EstimationThe model can accurately quantify and estimate the size, distance, and other parameters of objects.
• Spatial Relationship UnderstandingThe model can understand the relative positions, orientations, and spatial layouts of objects.
• Perspective ShiftIt supports processing information changes when viewing the same scene from different perspectives and inferring the impact of perspective changes.
• Spatial Reconstruction and DeformationTo understand the three-dimensional structure of an object and maintain spatial awareness after deformation or reconstruction.
• Spatial reasoningLogical reasoning based on spatial information, such as determining the direction of an object's movement or changes in spatial layout.
• Multimodal fusionBy combining multiple modalities of data, such as images and text, we can improve our ability to understand complex spatial scenes.

The technical principles of SenseNova-SI

• Scale effectThe core reason for the leap in SenseNova-SI performance is that the model's spatial cognitive ability can be significantly improved by training with large-scale, high-quality spatial data.
• Systematic training methodsSenseTime proposed a spatial capability classification system, based on which it expanded the data scale and adopted a systematic training method to enable the model to achieve consistent improvement across multiple spatial intelligence dimensions.
• Multimodal fusion architectureBased on infrastructure such as InternVL, SenseNova-SI can effectively integrate image and text information, improving its ability to understand complex scenes.