Google Gemini enables Interactions API: Refactoring agents to end 'ghost bills' According to Beating monitoring, Google has officially set the Interactions API as the default interface in Gemini API and Google AI Studio, and the original interface generateContent has entered a maintenance period. The new interface will package multiple rounds of conversations, model thinking, tool calls, and execution results, and store them as Interaction resources on the server side. The most direct benefit is saving money and effort. Due to the automatic saving of conversation status on the server side, during multiple rounds of chatting, the client only needs to send the interaction ID from the previous step, and Google can automatically reuse and improve cache hit rates, significantly reducing token fees. For time-consuming tasks such as code sandbox running or deep search, developers can enable background mode to run the work asynchronously in the background. In addition, the intermediate steps such as model thinking and tool calling are fully visible, making it easy to troubleshoot bugs in real-time during the operation of the intelligent agent. The underlying refactoring directly solved the billing crisis that erupted in early June. At that time, after manually deleting the cache, due to cloud synchronization delay, the logged out resources were still idle in the background, resulting in some accounts being abnormally deducted nearly 27000 yuan within 12 hours. In addition, the model is prone to getting stuck in a mental dead loop during online retrieval, and even in a "zero output" state without any response, it still charges at full 64000 tokens. The Interactions API replaces manual deletion with server hosted 'implicit caching' and allows developers to monitor and abort thought steps, providing a more secure billing defense. Due to the fact that Google's future cutting-edge models and intelligent agent features (such as Deep Research) will only support the Interactions API, it is recommended that developers complete the adaptation as soon as possible. [Original link]