FunctionGemma 270M Model (blog.google)

FunctionGemma, a new specialized version of the Gemma 3 270M model, has just been announced, enhancing the capabilities of AI agents by adding native function calling for on-device applications. This shift is pivotal for the AI/ML community as it signifies a move from passive conversational interfaces to active agents that can execute complex commands locally, automating tasks such as setting reminders and managing device settings. With over 300 million downloads and increasing demands from developers, FunctionGemma is engineered to run efficiently on edge devices while ensuring data privacy. The model allows developers to create customized, fast, and reliable local agents that translate natural language into executable actions. Its unique features include a unified approach for interacting with both users and systems, and a strong foundation for fine-tuning, which significantly boosts reliability and accuracy. FunctionGemma's small footprint, support for diverse tools, and emphasis on local-first deployment make it an attractive choice for applications with defined APIs and a need for instant responses. This advancement in AI toolkits opens new possibilities for user experiences, as highlighted in demonstrations within the Google AI Edge Gallery, where users can engage with interactive applications powered by this powerful model.