Google Gemini Upload Limits: File Caps, Size Ceilings & Large-Corpus Solutions

The Google Gemini upload limit restricts users to attaching a maximum of 10 files per prompt, with individual file sizes capped at 100MB and PDF documents limited to 48 pages on free accounts. Google documents that Gemini Apps accept up to 10 files in a single prompt, with non-video files capped at 100 MB and video files capped at 2 GB. Whether you submit brief technical notes, research PDFs, or complex spreadsheets, the web interface rejects any individual document larger than 100 MB or any prompt containing an eleventh file.

When users ask how many files can you upload to Gemini, Google restricts Gemini Apps uploads to 10 files in a single prompt across desktop web browsers, Android apps, and iOS clients. Under the official Gemini file upload size cap, Google restricts Gemini Apps non-video uploads to 100 MB per file, encompassing text documents, spreadsheets, presentations, and PDFs.

Google enforces distinct upload constraints depending on the access tier, interface, and protocol:

• Gemini Web Interface (Free Tier): Enforces the standard Gemini prompt upload limit of 10 files per prompt. Documents, spreadsheets, and text files are capped at 100 MB per file. Video files can reach up to 2 GB with a maximum duration of 5 minutes. Audio files are limited to 10 minutes.
• Gemini Advanced (Google AI Pro and Google AI Ultra Plans): Retains the standard 10-file attachment limit and 100 MB document size cap. The paid subscription increases video duration allowances to 1 hour and audio analysis to 3 hours.
• Google Workspace Accounts (Work and School): File uploads follow consumer size limits but are governed by organizational access policies. Connecting to Google Drive requires the Workspace administrator to enable Gemini Apps access.
• Gemini Files API (Developer Tooling): Intended for programmatic multimodal ingestion in Google AI Studio and Vertex AI. Supports files up to 2 GB each with a project storage limit of 20 GB. Uploaded files persist for 48 hours before automatic deletion.
• Inline API Payloads: Passing file bytes directly as base64-encoded strings inside prompt requests imposes a 100 MB total payload ceiling per API call (and 50 MB for PDF documents).

The following comparison table outlines the documented Gemini upload limits across environments as of October 2026:

Specific constraints apply when uploading compressed archives or project directories. Uploading a folder through the web interface unpacks its contents, with every individual file counting toward the 10-file maximum. If a folder holds 15 small text files, the upload fails. Standard ZIP archives are capped at 100 MB, cannot contain audio or video files, and cannot hold more than 10 internal items. Dedicated code folder uploads and connected GitHub repositories allow up to 5,000 files with a combined maximum archive size of 100 MB.

• Box File Size Limit: Upload Caps by Plan and Large-Corpus AI SearchBox documents an individual file upload limit across account plans, ranging from 250 MB on free accounts to 5 GB on...
• Google NotebookLM File Size Limit: Source Caps and Large-Corpus StorageGoogle NotebookLM restricts individual uploaded files to 200MB and 500,000 words across 50 sources on its free tier....
• Notion File Upload Limits: Size Caps by Plan and Large-File WorkaroundsNotion limits individual file uploads to 5MB on the Free plan and provides unlimited file uploads with a 5GB maximum...
• OneDrive File Size Limit: Upload Caps, Sync Bottlenecks, and AI Large-Corpus WorkspacesThe OneDrive file size limit is 250 GB per individual file across all account types, including Personal, Business, and...
• ChatGPT CSV Upload Limit: File Size Caps, Row Limits, and Memory WorkaroundsThe ChatGPT CSV upload limit allows raw files up to 512MB, but the Python sandbox frequently crashes on dataframes...
• ChatGPT Projects File Size Limit: Document Caps and Large-Corpus SearchThe ChatGPT Projects file size limit enforces a 512MB and 2 million token cap per text document, a 50MB ceiling on...

More on this subject: Agent File and Document Workflows (346 guides)

Clearing the 100 MB file size threshold does not guarantee that Gemini can process your document. File size measures disk storage in megabytes, whereas model comprehension is constrained by context window token capacity. A dense 40 MB text export can contain several million words, exceeding the processing capacity of any frontier model. Conversely, an 80 MB PDF containing scanned diagrams might contain only a few thousand words but will consume visual tokens for each rendered page.

Competitors frequently confuse Gemini's 1 million to 2 million token context window with prompt file upload caps, failing to explain that web UI upload limits trigger long before token exhaustion occurs. Gemini models feature extensive context windows, but those windows remain absolute boundaries:

• Free Web Tier: Features a 32,768 token context window, which translates to approximately 48 pages of standard text.
• Google AI Plus: Expands the context window to 128,000 tokens, processing roughly 190 pages.
• Deep Think Mode: Operates within a specialized 192,000 token context window, processing approximately 290 pages.
• Google AI Pro and Ultra: Provides a 1,000,000 token context window, capable of reading up to 1,500 pages of text or 30,000 lines of code.

When using the web interface, users encounter three specific failure modes and error warnings related to file processing:

• Your uploads may be too large for the best results: This warning appears when the attached document approaches the effective context limit. When prompted with massive documents, the model can miss connections or omit details scattered across distant sections.
• Delete data to upload file: Gemini Apps maintain an internal storage quota for data saved from previous chat activity, including uploaded files and generated artifacts. This storage is completely independent of Google Drive or Google Workspace storage limits. Users with terabytes of free Google Drive space still hit this wall if their Gemini chat history fills up. Resolving it requires deleting past Gemini chats or custom Gems.
• You have reached your limit for chats with files: Google enforces rolling capacity limits on how frequently users can upload and analyze files within a specific timeframe. Google publishes no fixed daily count or hourly quota, stating only that the limit resets after some time. Upgrading to paid tiers increases these rolling allowances but does not remove the cap.

For developers building automated pipelines, the Gemini web interface is not a practical ingestion layer. The 10-file cap and manual drag-and-drop workflow break down in production environments. To support larger payloads, Google provides the Files API for Google AI Studio and Vertex AI.

As documented in Google's official developer specifications, the Files API lets you store up to 20 GB of files per project, with a per-file maximum size of 2 GB. Files are stored for 48 hours before automated deletion. Google directs developers to use the Files API whenever a total request payload exceeds 100 MB, or whenever a single PDF document exceeds 50 MB.

The following Python example demonstrates uploading a large document to the Files API and querying it using the official Google GenAI SDK:

import os
from google import genai
from google.genai import types

client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])

uploaded_file = client.files.upload(
    file="audit_report.pdf",
    config=types.UploadFileConfig(
        display_name="Financial Audit Report",
        mime_type="application/pdf"
    )
)

print(f"File uploaded successfully: {uploaded_file.name}")
print(f"File URI: {uploaded_file.uri}")

response = client.models.generate_content(
    model="gemini-2.0-flash",
    contents=[
        uploaded_file,
        "Extract all vendor cost variances and list them in a Markdown table."
    ]
)

print(response.text)

client.files.delete(name=uploaded_file.name)

While the Files API solves the 100 MB per-file limitation of the web UI, it introduces operational constraints for production teams:

1. Ephemeral Lifecycle: Files vanish after 48 hours. If your application references technical documentation, standard operating procedures, or historical contracts, your code must continually re-upload files. This causes repeated upload latency, excessive bandwidth consumption, and duplicated processing.
2. No Persistent Team Workspaces: The Files API is an ephemeral staging bucket tied to an API key. It lacks shared team workspaces, version histories, access management, and collaborative review mechanisms.
3. Download Restrictions: Google allows developers to retrieve metadata about uploaded files, but user-uploaded file bytes cannot be downloaded back through the API.

Query large document libraries without prompt upload limits

Store your multi-gigabyte document archives in Fastio workspaces, enable automated semantic search, and connect your AI agents through remote MCP endpoints. Monthly plans include a 14-day free trial, which requires a credit card. Plans are Starter at $9.99/mo, Business at $49.99/mo, and Enterprise at $199.99/mo.

Start 14-Day Trial

Evaluating file upload constraints across frontier AI platforms reveals notable differences in how vendors balance file size, file count, and context persistence. Understanding these differences helps technical teams select the right assistant for their document processing tasks.

• Claude (Anthropic): In a standard Claude chat, users can attach up to 20 files per prompt, with each file capped at 500 MB. In Claude Projects, individual files are limited to 30 MB each, but there is no fixed file-count cap. A project can hold hundreds of files as long as the combined project knowledge fits within Claude's 200,000-token context window.
• ChatGPT (OpenAI): Supports file uploads up to 512 MB per document. Free tier accounts can upload only 3 files per day, while paid Plus, Team, and Enterprise accounts permit 10 files per message and up to 80 files every 3 hours. Spreadsheets are capped at 50 MB, and CSV files cannot exceed 10,000 rows.
• Google Gemini: Restricts web users to 10 files per prompt and 100 MB per document, while offering up to 2 GB for video files. Effective reading capacity ranges from 48 pages on free accounts to 1,500 pages on AI Pro plans.

The comparison table below details the practical limits across each platform:

Each platform presents a distinct tradeoff. Gemini leads the market in single-file video ingestion and token context volume on paid plans, but imposes the tightest file-count ceiling per prompt (10 files) and a modest 100 MB document cap. Claude offers generous per-file caps in chat (500 MB) and flexible project organization, but enforces a 200,000-token ceiling. ChatGPT provides a balanced middle ground for individual documents but restricts daily free usage.

When analyzing a corporate repository containing thousands of technical specifications, legal contracts, or customer support archives, dragging 10 files at a time into a chat prompt is unsustainable. Slicing PDFs into 48-page fragments or re-uploading entire directories to ephemeral API buckets every 48 hours wastes engineering hours.

Teams facing these limits typically evaluate three architectures:

1. Local storage and self-hosted vector databases: Running local vector stores like Chroma or Qdrant paired with local file systems allows semantic search, but requires continuous pipeline maintenance, embedding updates, and dedicated host resources.
2. Raw cloud object storage: Storing document collections in Amazon S3 or Google Cloud Storage resolves raw capacity ceilings, but leaves retrieval unhandled; engineering teams must manually construct chunking, indexing, and context-retrieval middleware before an LLM can query the archive.
3. Intelligent workspaces: Platforms that pair persistent cloud workspaces with automated hybrid search and open protocol connectivity bridge the gap between static file storage and conversational retrieval.

Intelligent workspaces solve the architectural bottleneck of prompt limits by decoupling document storage from prompt payloads. Instead of attaching raw file bytes directly to every conversation, teams store their corpus in a persistent, organization-owned workspace and connect their AI assistant through the Model Context Protocol (MCP).

Fastio workspaces provide an intelligent platform designed specifically for teams using storage for agents and handling document-heavy operations:

• Direct and Scheduled Ingestion: Workspaces ingest multi-gigabyte document collections through chunked uploads supporting files up to 25 GB on Starter, 50 GB on Business, and 100 GB on Enterprise plans. Alternatively, Cloud Sync connects Google Drive, Dropbox, Box, and OneDrive folders to a workspace, one-way or two-way, on a schedule or on demand, without routing data through local developer machines.
• Automated Indexing with Intelligence Mode: Once Intelligence Mode is enabled on a workspace, all uploaded and synced documents are automatically indexed for hybrid search, combining full-text keyword indexing, semantic meaning, and metadata retrieval. Teams avoid setting up external vector databases, managing embedding models, or maintaining chunking pipelines.
• Remote Model Context Protocol (MCP) Connectivity: AI assistants connect to Fastio workspaces over Streamable HTTP using client-specific endpoints. Claude applications and general tools connect at https://mcp.fast.io/mcp/tools. ChatGPT and Codex connect via the Fastio plugin, with custom servers using https://mcp.fast.io/mcp/operations. Coding agents like Gemini CLI, Cursor, Claude Code, and Cline connect at https://mcp.fast.io/mcp/code. Setup procedures are documented at https://mcp.fast.io/docs.
• Precision Retrieval with Source Citations: Instead of forcing hundreds of megabytes of raw files into the model context window, the AI assistant queries the workspace via MCP tools. The search engine retrieves only the relevant passages and presents answers backed by citations to specific workspace documents. This approach keeps prompts lean, avoids file upload ceilings, and eliminates token bloat.
• Structured Data Extraction with Metadata Views: For teams that need structured records rather than conversational summaries, Metadata Views extracts typed fields (Text, Integer, Decimal, Boolean, URL, JSON, Date & Time) across document collections into sortable, filterable tables without manual OCR templates.
• Governance and Access Control: Role-based access permissions are established at the organization, workspace, and share levels. A detailed activity log records every modification, file addition, deletion, and share creation, maintaining an accountable audit record across team members and autonomous agents. Review available pricing plans to match storage volume with your team's document requirements.

For developer environments running Gemini CLI or compatible forks like Qwen Code, connect to the workspace by adding an entry to settings.json under mcpServers:

{
  "mcpServers": {
    "fastio": {
      "httpUrl": "https://mcp.fast.io/mcp/code"
    }
  }
}

In these clients, httpUrl selects the Streamable HTTP transport (using url inadvertently selects SSE, which is unsupported). Interactive clients sign in with OAuth in the browser to grant access to the appropriate workspaces.