Google releases SigLIP 2: Multilingual Vision-Language Encoder

AI Impact Summary

Google released SigLIP 2, a new family of multilingual vision-language encoders that significantly improves upon the original SigLIP model. Key advancements include the introduction of dynamic resolution (naflex) variants for handling varying aspect ratios and a larger ‘giant’ model (1B parameters) that demonstrates superior performance across core capabilities like zero-shot classification and image-text retrieval. These improvements position SigLIP 2 as a strong foundation for building more robust and adaptable Vision-Language Models (VLMs).