Google Unveils Agent Vision Feature: AI Image Comprehension Evolves from Static Analysis to Dynamic Exploration
2 week ago / Read about 0 minute
Author:小编   

On January 27, 2026, Google's DeepMind team rolled out the 'Agent Vision' feature for the Gemini 3 Flash model. This innovative feature emulates human visual processing via a closed-loop cycle of 'think-act-observe.' It empowers the model to dynamically explore and interpret visual data, thereby minimizing uncertainty and inaccuracies. In numerous tests, this approach has led to a 5-10% enhancement in the quality of visual understanding.