SenseTime has open-sourced the SenseNova-MARS AI model that combines image and text searching with smart reasoning to solve tough visual puzzles.
The tool lets AI think like a detective by zooming into photos, hunting for facts online, and piecing together answers step by step.
Imagine showing the AI a blurry photo of a rare bird in a forest. Instead of guessing, SenseNova-MARS searches the web for similar images and facts, crops in on tiny details such as feathers or markings, and reasons through clues over multiple steps to identify it correctly.
It uses tools for text search, image search and precise image cropping, all powered by reinforcement learning, a training method that rewards smart decisions, much like teaching a child through trial and error.
As the first open-source AI of its kind to handle dynamic video reasoning alongside images and text, it is claimed to even top closed models such as Gemini-3-Pro and GPT-5.2 on tests for search accuracy (74.3 on MMSearch) and detailed image analysis.
Available in 8B and 32B sizes on GitHub and Hugging Face, developers worldwide can now build smarter apps for robotics, self-driving cars, or everyday photo searches without starting from scratch.
Entelechy Asia 