On January 29, 2026, SenseTime officially announced the open-source release of its multimodal autonomous reasoning model SenseNova-MARS, along with two versions: 8B and 32B. The launch of this model marks a critical step forward in the field of multimodal large models for autonomous reasoning.
Technical Breakthrough: First Agentic VLM Model
SenseNova-MARS has achieved significant innovation in its technical architecture, making it the first Agentic VLM (Agentive Vision-Language Model) in the industry that integrates dynamic visual reasoning with text-image search.
Autonomous Reasoning: The model not only understands image content but also possesses autonomous planning and reasoning capabilities similar to those of an agent.
Deep Integration: By integrating real-time search capabilities into the visual understanding process, the model can handle complex visual tasks that require external knowledge support.
Industry Impact and Significance
SenseTime's decision to open-source both versions aims to provide global developers with more flexible research tools:
8B Version: Balances performance and efficiency, suitable for deployment on edge devices or in environments with limited computing power.
32B Version: Offers stronger logical reasoning capabilities, meeting the needs of complex industry applications.
