Google supercharges Gemini API with massive file limits and cloud storage integration

Google is continuing to break down the barriers for developers building multimodal AI applications with a significant update to the Gemini API. While early versions of the API focused on getting models into the hands of developers quickly, the latest enhancements address a major pain point: data ingestion at scale.

By increasing file size limits and introducing direct support for external cloud storage, Google is making it significantly easier to move from small-scale prototypes to production-ready software.

The headline change is a massive jump in inline file size limits, moving from a 20MB cap to a much more substantial 100MB. For developers working with high-resolution images, lengthy audio clips, or complex PDF documents, this five-fold increase simplifies the development process by allowing larger payloads to be sent directly within the API request without the need for intermediate storage.

Breaking the storage bottleneck

Beyond the increased capacity, Google is introducing two new methods for handling data that resides outside of the Gemini environment. Previously, large assets like long-form video or massive document libraries had to be uploaded to the Gemini Files API, where they were only stored for 48 hours; this was a hurdle for enterprise-grade applications that rely on persistent data.

To solve this, the Gemini API now supports direct object registration from Google Cloud Storage (GCS). This means if your data is already sitting in a GCS bucket, you no longer need to move bytes around. Developers can register these files once and reference them across multiple requests, which drastically reduces the time and cost associated with data transfer.

Furthermore, the API now supports HTTPS and Signed URLs. This expansion allows Gemini to fetch content directly from the web or private storage across other cloud providers like AWS S3 or Azure Blob Storage. By using pre-signed URLs, the Gemini API securely fetches the content during processing, eliminating the need for developers to download content to their own backends just to forward it to the model.

Moving toward seamless multimodal production

These updates are powered by the latest Gemini 3 models, which are specifically designed to handle the complex reasoning required for large-scale multimodal inputs. Whether it’s analyzing a 45-minute video stored in an S3 bucket or summarizing a hundred-page technical manual via a signed URL, the underlying architecture is now built to treat external data as if it were local.

For the developer community, this shift signifies Google’s commitment to making the Gemini ecosystem as flexible as possible. By meeting data where it lives, rather than forcing it into a proprietary silo, Google is positioning the Gemini API as a highly versatile engine for the next generation of AI-driven tools. These features are available today in the latest GenAI SDKs, allowing developers to start scaling their applications immediately.