Gemini API Backward Incompatible Update: On June 8th, all legacy SDKs threw errors, while the new endpoint enabled multi-modal simultaneous output.

According to Perceptual Beating monitoring, Google has announced that the Gemini Interactions API will undergo a breaking change. The new version of the interface will become the default on May 26, and the old version will be deprecated on June 8. At that time, requests using the Python or JavaScript 1.x.x SDK will throw errors directly.

Developers need to modify two aspects of their code logic:
First, how to extract the response content. The original outputs array has been deprecated, and content retrieval has been entirely moved to the more granular steps array. Whether reading text generated by the model or the result of a function call, developers must rewrite their parsing code according to the new structure.
Second, how to configure the output format. The previously scattered MIME type declarations and image generation parameters (`image_config`) have now been consolidated into the response_format parameter. Standardizing the format has introduced a new capability: developers can pass an array to the response_format, allowing the model to output text and multimodal results such as audio in a single request.

SDK users can now upgrade to version 2.0.0 or above to automatically switch over. REST API users can add the Api-Revision: 2026-05-20 request header to proactively test the new version; once the new version becomes the default on May 26, if the code is not yet updated, the request header can briefly revert to 2026-05-07 until the complete shutdown on June 8.