Gemini API request storage: what changed for new paid projects
Google changed the default for new paid Gemini API projects on October 26, 2026: Generate Content requests are now stored unless the project or request opts out. Existing projects retain their previous setting. The change covers generateContent, streamGenerateContent and batchGenerateContent, according to the official changelog.
This is an API configuration change, not an announcement about the consumer Gemini app. It matters when a team creates a new project expecting the same logging behavior as an older deployment.
Check the project and the request separately
The logging guide says a request’s store value overrides its project default. Set store=false for a request or change Generate Content logging in AI Studio’s Logs settings. The Interactions API has a separate control. Disabling its storage affects automatic conversation history.
| Generate Content SDK | Minimum version for request-level store |
|---|---|
| Python google-genai | 2.33.0 |
| JavaScript @google/genai | 2.31.0 |
| Java google-genai | 1.79.0 |
| Go google.golang.org/genai | 1.76.0 |
Logs default to 55 days before expiry and deletion marking, with 7, 14 or 28 day alternatives. Saving logs as datasets changes that lifecycle: datasets do not expire. Developers can also choose to share datasets with Google for research and development. Logging also has exclusions, including video, GIF and PDF inputs, so an empty dashboard is not proof that a request was never processed.
Storage is not the same as model training
Google’s Gemini API terms say Paid Services prompts and responses are not used to improve its products. The terms separately allow limited logging for safety enforcement and required disclosures. Turning off developer-facing request storage should therefore not be described as a guarantee of zero retention across every Google system.
A practical deployment review
Compare development, staging and production rather than assuming their defaults match. Record the project creation date, API route, SDK version and intended retention policy. Then use a harmless test prompt to check the behavior your application actually sends, including wrappers that may supply their own request options.
Keep exported evaluation datasets in that review too. Turning a request log into a reusable test fixture is a separate decision from keeping short-lived debugging records. ByteForward has reviewed the documentation, not changed a customer’s project settings or tested these controls in a live application.
Featured image: AI-generated editorial illustration by ByteForward with the official wordmark from the Gemini website. Gemini is a trademark of Google LLC. This article is not affiliated with or endorsed by Google. The archive is conceptual, not a product screenshot.
