Analysis
Shangtang's SenseNova-MARS, a new open-source multimodal autonomous reasoning model, has made waves by surpassing Gemini-3 Pro in key benchmark tests. This achievement highlights the rapid advancement of open-source AI, offering developers and users a powerful new tool for complex tasks involving visual understanding and information retrieval.
Key Takeaways
- •SenseNova-MARS is an open-source model that outperforms Gemini-3 Pro and GPT-5.2 in multimodal reasoning and search.
- •It's the first Agentic VLM model supporting dynamic visual reasoning and deep integration of image and text search.
- •The model, code, and data are fully open-sourced on Hugging Face and Github, promoting accessibility and collaboration.
Reference / Citation
View Original"Today, Shangtang officially open-sourced the multimodal autonomous reasoning model SenseNova-MARS (8B/32B dual versions), which surpassed Gemini-3-Pro (69.06 points) and GPT-5.2 (67.64 points) with 69.74 points in the core benchmark tests for multimodal search and reasoning."