Google announces Gemma 2, a 27B-parameter open model, and PaliGemma, a vision-language model, at Google I/O 2024.

Google's annual Google I/O 2024 developer conference announced Gemma 2, a 27B-parameter version of the open model, launching in June. The company also revealed PaliGemma, the first vision-language model in the Gemma family for image captioning, image labeling, and visual Q&A use cases. The current Gemma models, launched earlier this year, only had 2B and 7B parameter versions.