Google’s latest open-source multimodal model, Gemma 3 27B, supports vision-language input and text outputs, handles context windows up to 128k tokens, and understands over 140 languages. Offers improved math, reasoning, and chat capabilities, including structured outputs and function calling.