Starting today, developers using Google’s Gemini API and Google AI Studio will have the option to enhance their AI services and bots with real-time data from Google Search, offering users more accurate and up-to-date responses. This feature, known as "grounding," allows the Gemini API to pull in verified, dynamic information, reducing the risk of hallucinations — when a model provides incorrect information.
While AI Studio users can test grounding for free, paid Gemini API users will be charged $35 per 1,000 grounded queries. Grounding is activated by a toggle and adjusted through the “dynamic retrieval” setting, allowing developers to decide the frequency and context in which grounding pulls in data from Google Search. For instance, developers might prefer grounding on recent events or for all queries to gain richer detail.
AI Studio now offers a "compare mode" that lets developers directly see the differences between grounded results and those generated solely by the model’s pre-existing knowledge, showcasing the improvements made through grounding. Google’s Logan Kilpatrick highlights that displaying source links with grounded results is a requirement, ensuring that original publishers get visibility and enabling users to verify the information themselves.
AI Studio has evolved beyond a simple prompt testing tool; it now provides developers with code generation to integrate Gemini models into real-world applications. The streamlined experience in AI Studio allows developers to experiment with models and quickly bring code into production, making it easier to create AI-enhanced applications.
