Gemini: file upload limits and supported formats

The fact that Gemini handles file uploads is interesting to anyone who relies on documents, spreadsheets, media files, or code within interactive workflows. The system applies different size limits, token conversion rules, and integration constraints depending on the file type, and these directly affect usability in both free and enterprise tiers.

Gemini applies file size and format limits depending on media type.

Gemini defines strict thresholds for different file categories to ensure stability and predictable context usage. Documents such as PDF, DOCX, TXT, and HTML are supported up to 30 MB or 2,000 pages, with text extracted into tokens. Spreadsheets can reach 20 MB or 1,000,000 cells, and presentations such as PPTX or Google Slides exports are capped at 35 MB and 500 slides.

Media types follow a different set of rules: images (JPG, PNG, WEBP, SVG) are limited to 15 MB or 24 megapixels per file with a maximum of three per prompt, while audio (WAV, MP3, FLAC) accepts 40 MB or about 20 minutes. Video files remain in beta, capped at 200 MB or 30 minutes, processed through key-frame sampling at one frame per second. Code files are accepted up to 10 MB or 25,000 lines, and compressed ZIP archives containing images or frames can be uploaded up to 150 MB.

Media type	Formats	Max size	Special limits
Documents	PDF, DOCX, TXT, HTML	30 MB / 2,000 pages	Extracted as text tokens
Spreadsheets	XLSX, CSV	20 MB / 1,000,000 cells	Formulas flattened
Presentations	PPTX, Slides export	35 MB / 500 slides	Speaker notes included
Images	JPG, PNG, WEBP, SVG	15 MB / 24 MP	Up to 3 per prompt
Audio	WAV, MP3, FLAC	40 MB / 20 min	Auto-transcribed
Video	MP4, MOV	200 MB / 30 min	Beta, key-frame analysis
Code	py, js, java, cpp, html	10 MB / 25,000 lines	Parsed into syntax tree
ZIP (images/frames)	ZIP	150 MB / 100 images	Each frame parsed separately

Upload workflows differ between Gemini web, mobile, and Workspace integrations.

File handling varies across Gemini entry points. In the web chat interface, uploads are capped at 20 MB each with a total of 100 MB per thread. The mobile app enforces a stricter 15 MB per file and 50 MB per day. In Docs and Slides, Gemini can process any Drive-linked file up to 100 MB, respecting existing sensitivity labels. In NotebookLM, corpus uploads allow 100 MB per document and up to 2 GB per notebook.

These variations matter because the same file may be usable in Workspace but rejected in chat. The reliance on Drive integration means that confidentiality labels such as “Internal” or “Confidential” affect not only sharing but also output policies, with restricted summaries or redactions applied automatically.

Surface	Upload cap	Batch/session limit
Web chat	20 MB/file	100 MB/thread
Mobile app	15 MB/file	50 MB/day
Docs sidebar	100 MB/file	Drive quota
NotebookLM v2	100 MB/file	2 GB/notebook

File size converts into tokens and consumes the context window.

Every upload is internally tokenized, which impacts the usable context window. Text follows the rule of 1 KB ≈ 4 tokens, images count as 500 tokens each, audio at about 1 token per second, and video frames at 120 tokens per frame. The free tier supports 128,000 tokens, while enterprise and API users access up to 1,000,000 tokens.

This conversion model means that large PDFs, long audio recordings, or frame-heavy video extractions can quickly consume available context. When limits are exceeded, Gemini throws a context overflow error that guides users to chunk or trim the input before reprocessing.

File type	Token ratio	Free tier	Enterprise
Text	1 KB ≈ 4 tokens	128,000	1,000,000
Images	500 tokens/image	3 images	10 images
Audio	1 sec ≈ 1 token	5 min	15 min
Video frames	120 tokens/frame	150 frames	500 frames

Enterprise governance features restrict storage and usage of uploaded files.

Gemini applies governance rules around storage, training exclusion, and audit logging. The no-train toggle ensures uploaded files are excluded from future model training, which is the default for enterprise users. Region locks keep EU files in European data centres, while audit logs capture SHA-256 hashes, file size, user ID, and latency for every upload. Files undergo virus scanning and DLP checks, blocking executables or sensitive PCI data from being processed.

These features ensure compliance with corporate security policies and international regulations while still allowing productivity use cases.

Rate limits depend on subscription level.

Upload frequency and daily storage allowance vary significantly across plans. Free plans are capped at 2 uploads per minute and 200 MB per day. Advanced subscriptions allow 10 per minute and 5 GB daily, while Enterprise and Ultra users may raise quotas up to 30 per minute and 20 GB daily by request.

Plan	Uploads/minute	Daily quota
Free	2	200 MB
Advanced	10	5 GB
Enterprise/Ultra	30 (scalable)	20 GB

Known limitations still require work-arounds.

Despite broad file support, several practical issues remain. Uploading very large videos leads to 413 payload errors, solved by segmenting into smaller files. Legacy .xls spreadsheets cause parsing errors, making conversion to CSV or XLSX necessary. Images with faces may be rejected in sensitive workspaces unless rights verification is applied. Processing 4K video slows frame sampling, and downscaling to 1080p is recommended.

These work-arounds highlight the balance between capacity and performance, with users adapting to Gemini’s technical boundaries.

Roadmap promises chunking, Drive folder uploads, and code execution.

Upcoming changes include an automatic PDF chunker that groups pages by semantics, Drive folder uploads with hierarchical context tagging, and a sandbox for code execution where uploaded scripts can be tested inline. These additions are expected to make file uploads more flexible, especially for research and enterprise development workflows.

With file handling becoming a central element of productivity assistants, Gemini’s evolving limits and integration rules define how effectively documents, spreadsheets, media, and code can be used in daily work.

____________

FOLLOW US FOR MORE.

DATA STUDIOS

datastudios.org