Google has unveiled PaliGemma 2, its latest open-source family of vision-language models, combining SigLIP for vision encoding and Gemma 2 for text decoding. Available today in parameter sizes of 3 billion, 10 billion, and 28 billion—or 2 billion, 9 billion, and 27 billion—PaliGemma 2 supports image resolutions of 224x224, 448x448, and 896x896 pixels. Designed for easy fine-tuning across various vision-language tasks, the models offer advanced features like generating detailed, context-sensitive image descriptions and improved performance in applications such as chemical formula recognition, music score interpretation, spatial reasoning, and chest X-ray report generation. PaliGemma 2 is accessible on platforms like Kaggle Models and HuggingFace Transformers, facilitating widespread adoption by developers and researchers.