In a major update for AI developers, Google officially announced on January 12, 2026, a significant expansion of data ingestion capabilities for the Gemini API. The headline change is a 5x increase in the maximum size for inline file data, moving the limit from 20MB to 100MB.
This update is designed to eliminate friction for developers building multimodal applications that require real-time processing of large documents, high-resolution images, and short video clips.
Breaking the 20MB Barrier: New File Input Methods
Previously, developers handling files larger than 20MB were forced to use the ephemeral Files API, which required a separate upload step and stored data for only 48 hours. The new update introduces three distinct ways to bring large datasets into Gemini 3 and Gemini 2.5 models.
1. Increased Inline Limit (100MB)
For “speed and simplicity,” developers can now send up to 100MB of base64 encoded data directly within the request payload. This is ideal for:
- Real-time applications where low latency is critical.
- Prototyping without managing external storage.
- Complex PDF analysis (though PDFs remain capped at 50MB for certain optimized processing paths).
2. External HTTP & Signed URLs
Gemini API now natively supports fetching data from public URLs or pre-signed URLs from providers like AWS S3, Azure Blob Storage, or private web servers.
- No Re-uploads: The API fetches content during processing, saving bandwidth and backend overhead.
- Scale: This method supports files up to 100MB per request, perfectly suited for production apps with persistent data.
3. Google Cloud Storage (GCS) Integration
For enterprise users, Google has introduced GCS Object Registration. Developers can now point the Gemini API directly to their existing Google Cloud buckets.
- Large File Support: Handles files up to 2GB (standard for video and massive archives).
- Persistence: Access can be granted for up to 30 days without re-uploading the data.
Comparing Gemini API Input Methods (2026)
| Method | Best For | Max Size | Persistence |
| Inline Data | Quick testing, real-time apps | 100 MB | None (Sent per request) |
| External URLs | Data on AWS, Azure, or Web | 100 MB | None (Fetched per request) |
| Files API | Reusing files multiple times | 2 GB | 48 Hours |
| GCS Registration | Enterprise data in Google Cloud | 2 GB | Up to 30 Days |
Why This Matters for Developers
This update effectively turns the Gemini API into a “bring-your-own-data” powerhouse.14 By allowing the model to pull data directly from where it lives, Google is removing the “ephemeral storage bottleneck” that plagued early multimodal development.
Key Use Cases Enabled by the 100MB Limit:
- Deep Document Research: Analyzing massive, 500-page legal or medical PDFs without manual splitting.
- High-Fidelity Audio: Processing long-form audio files (up to 3 hours on Pro models) in a single API call.
- Native Video Analysis: Using the Veo 3 or Gemini 3 Pro models to summarize 4K video clips directly from a cloud bucket.
Getting Started
These new methods are available immediately across all latest Gemini SDKs. Developers can begin using the 100MB inline limit in Google AI Studio and Vertex AI starting today
