Back to ModelsGoogle
Gemma 3
Open source4 variants
Google's tiny 1B Gemma 3 instruct model — designed for on-device inference with solid multilingual instruction-following.
33K tokensFree / Open weightsTransformerGemma Terms
BFCL v4
gemma-3-1b-it
7.2%gemma-3-27b-it
29.5%gemma-3-4b-it
19.6%Gemma-3-1b-it Prompt — rank #109
base270M
Google's tiny 270M Gemma 3 base model — the smallest of the Gemma 3 family, built for ultra-low-resource on-device inference.
8K tokensFree / Open weightsTransformer
instruct27B
Google's 27B Gemma 3 multimodal instruct model — strongest of the Gemma 3 family with vision understanding and 128K context.
131K tokensFree / Open weightsimage
instruct4B
Google's compact 4B Gemma 3 multimodal instruct model — strong vision and text capabilities in an efficient on-device package.
131K tokensFree / Open weightsimage
