Read, Watch or Listen

Google unveiled Gemma 3 270M, a significantly smaller version of its generative AI model. Unlike larger models requiring powerful cloud servers, this version runs efficiently on smartphones and web browsers, prioritizing privacy and speed. Despite its 270 million parameters (compared to billions in larger models), it demonstrates strong instruction-following capabilities, scoring higher than other lightweight models on the IFEval benchmark. While performance doesn't match larger models, its efficiency and local processing make it ideal for various applications.

Prepared by Jonathan Pierce and reviewed by editorial team.