Google AI Studio file upload and reading limits 2026: 2 GB files, 50 MB PDFs

Updated: November 7, 2026

Files sent to Gemini through Google AI Studio and the Gemini API follow the limits of the Files API: up to 2 GB per file, 20 GB of storage per project, and automatic deletion after 48 hours, at no cost in every region where the Gemini API is available. PDFs have a tighter ceiling of 50 MB or 1,000 pages, and files sent inline with a request are capped at 100 MB per request or payload, according to Google's developer documentation as checked on November 7, 2026.

The consumer Gemini app uses a different rulebook, with up to 10 files per prompt, 100 MB per non-video file and 2 GB per video, so a file that uploads fine in one product can be rejected in the other. The table below collects the official limits for each way of sending a file, followed by the per-format rules for PDFs, images, video and audio, a token and cost example computed from Google's published rates, and the steps to upload and manage files.

........

Upload route	Size limit	How long it is kept	Best suited for
Inline data in the request	100 MB per request or payload (50 MB for PDFs)	Not stored, sent with each request	Small files and quick tests
Files API upload	2 GB per file, 20 GB per project	48 hours, then deleted automatically	Large files reused across several prompts
Google Cloud Storage URI registration	2 GB per file, no overall storage limit	Not stored, fetched per request	Files already kept in a Cloud Storage bucket
Public or signed external URL	15 MB per file	Not stored, fetched per request	Data you do not want to upload again
Gemini app (consumer)	10 files per prompt, 100 MB per file, 2 GB per video	Retention not published	Conversational analysis outside the API

........

··········

GOOGLE AI STUDIO FILES API LIMITS: SIZE, STORAGE AND 48-HOUR RETENTION.

The Files API is the route Google recommends whenever the total size of a request, including the prompt, system instructions and attached files, goes above 100 MB, and for any file that will be referenced by more than one request. Each uploaded file can reach 2 GB, and a single project can hold up to 20 GB at a time. Google states that files are stored for 48 hours and deleted automatically after that window, so a pipeline that runs every day has to upload its source files again or switch to a storage method that does not expire.

The documentation lists four operations: upload, get metadata, list and delete. A developer can remove a file before the 48 hours are up to free part of the 20 GB quota. Files that a user uploads cannot be downloaded back through the API; only files generated by a model, such as videos, can be retrieved with the download method. The Files API itself carries no charge, and the cost of a request comes from the tokens the uploaded content produces once a model reads it.

For files that already live in Google Cloud, the Gemini API can register a Cloud Storage URI instead of copying the bytes. The limit stays at 2 GB per file, with no overall storage cap, and the registration can stay valid for up to 30 days. Access requires granting the Gemini API service agent the Storage Object Viewer role on each bucket. External links work too: a public or pre-signed URL can point to a file of up to 15 MB, although Google notes that the Gemini 2.0 family of models does not support this method.

Google adds a caveat on every limit in this section: the effective ceiling can vary with the file type and with the model or tokenizer that processes the file. A file that fits the byte limit can still fail if its token count exceeds the context window of the model chosen in AI Studio.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣

GOOGLE AI STUDIO PDF AND DOCUMENT READING LIMITS.

PDFs get their own limit of 50 MB or 1,000 pages per file, and the same ceiling applies whether the PDF is sent inline or uploaded through the Files API. Gemini reads PDFs with full visual understanding, which means charts, tables, diagrams and page layout are part of what the model sees. Google counts each document page as 258 tokens, and native text extracted from the PDF is not billed separately; the page images are counted under the image modality.

Page images are resized before processing. Large pages are scaled down to a maximum of 3072 x 3072 pixels and small pages are scaled up to 768 x 768 pixels, with the aspect ratio preserved. Gemini 3 models add a media resolution setting with low, medium and high levels, which trades token usage for better recognition of small print. There is no fixed number of PDFs per request: several documents can go into a single prompt as long as their combined size and the text prompt stay within the model's context window.

Other text formats behave differently. Google lists TXT, Markdown, HTML and XML as formats that can be passed to the model, but they are extracted as plain text, so formatting, charts and diagrams in those files are lost. The document guide does not mention Word files or CSV files, which is why converting a Word report to PDF before uploading keeps its tables and figures readable. Our separate guide to Google AI Studio PDF reading and extraction covers extraction accuracy in more depth, and spreadsheets have their own page on Excel and CSV uploads in Google AI Studio.

··········

GOOGLE AI STUDIO IMAGE, VIDEO AND AUDIO FILE LIMITS.

Each media type has its own list of accepted formats and its own token rate, and those rates decide how much material fits into one prompt. The figures below come from Google's image, video and audio understanding pages for the Gemini API.

........

File type	Accepted formats	Per-prompt limit	Token rate
Images	PNG, JPEG, WEBP, HEIC, HEIF	Up to 3,600 image files per request	258 tokens per image up to 384 px; larger images split into 768 x 768 tiles of 258 tokens each
Video	MP4, MPEG, MOV (QuickTime), AVI, FLV, MPG, WEBM, WMV, 3GPP	Up to 3 hours at low resolution or 1 hour at high resolution on 1M-token models; up to 10 videos per request on Gemini 2.5 and later	About 100 tokens per second at default resolution, about 300 at high resolution
Audio	WAV, MP3, AIFF, AAC, OGG, FLAC, MPEG, M4A, L16, Opus, ALAW, MULAW, WEBM	Up to 9.5 hours of audio per prompt	32 tokens per second (1,920 tokens per minute)
YouTube links	Public videos only	Free tier: up to 8 hours of YouTube video per day; paid tier: no length-based limit	Same as video

........

Video is sampled at one frame per second by default, and Google suggests a lower rate for long, mostly static recordings and a higher one for fast action. Audio is downsampled to 16 Kbps before the model processes it. YouTube links are in preview and only public videos are accepted, so private and unlisted videos have to be downloaded and uploaded as files instead.

One inconsistency in Google's documentation affects images sent inline. The file input methods page, updated on October 7, 2026, gives 100 MB per request or payload for inline data, while the image understanding page still says inline image data limits the total request to 20 MB. Keeping inline image requests under 20 MB, or uploading larger image sets through the Files API, avoids depending on which figure the endpoint applies.

··········

[[ADPLUS_300x250_1]]

GEMINI APP UPLOAD LIMITS COMPARED WITH GOOGLE AI STUDIO.

The Gemini app on the web and on mobile applies separate rules, set out in the Gemini Apps Help Center. Up to 10 files can be uploaded in the same prompt, and every file inside an uploaded folder counts toward that number. Videos can be up to 2 GB each and every other supported file type up to 100 MB. ZIP archives can hold up to 10 files and 100 MB and cannot contain video or audio. A code folder or a GitHub repository can be added with up to 5,000 files and 100 MB in total.

Length limits depend on the subscription. By default, total video length is capped at 5 minutes and total audio length at 10 minutes. Upgrading to Google AI Pro or Google AI Ultra raises those totals to 1 hour of video and 3 hours of audio, and the same Help Center page lists a 1 million token context window for those plans. Uploads in the app are also subject to rolling usage limits over time. Files from Google Drive can be added with the Add from Drive option, which requires Keep Activity to be on and Google Workspace to be connected to the Gemini app.

........

Limit	Gemini app	Google AI Studio and Gemini API
Files per prompt	Up to 10	No fixed count, bounded by the context window (images up to 3,600)
Maximum size per file	100 MB, or 2 GB for video	2 GB through the Files API, 50 MB for PDFs
Video length	5 minutes, or 1 hour with Google AI Pro or Ultra	Up to 3 hours at low resolution on 1M-token models
Audio length	10 minutes, or 3 hours with Google AI Pro or Ultra	Up to 9.5 hours per prompt
Storage period	Not published in the Help Center	48 hours in the Files API, then deleted

........

The Gemini app suits occasional analysis of a document or a recording inside a conversation. AI Studio and the API are the better fit for long recordings, large media sets and automated workflows, at the price of managing the 48-hour expiry. Companies on Google Cloud have a third option in the Gemini Enterprise Agent Platform, whose documentation still sits at the Vertex AI address and lists its own per-model limits for documents and media; those pages could not be read in full during this update, so their figures are left out here.

··········

GOOGLE AI STUDIO FILE TOKEN COSTS WITH WORKED EXAMPLES.

Uploading is free, but every file turns into input tokens when a model reads it. The examples below apply Google's published token rates (258 tokens per PDF page, about 100 tokens per second of video at default resolution) to the standard paid-tier input prices listed on the Gemini API pricing page on November 7, 2026: Gemini 3.5 Flash-Lite at $0.30 per 1M input tokens, Gemini 3.8 Flash at $0.75 through December 31, 2026 and $1.50 from January 1, 2027, and Gemini 3.1 Pro Preview at $2.00 per 1M tokens for prompts up to 200,000 tokens and $4.00 above that. Gemini 3.5 Flash-Lite and Gemini 3.8 Flash also have a free tier where input is free of charge; Gemini 3.1 Pro Preview has none. The figures cover input only and exclude the prompt text and the model's output.

........

File	Input tokens	3.5 Flash-Lite	3.8 Flash (Dec 2026)	3.8 Flash (from Jan 2027)	3.1 Pro Preview
300-page PDF	77,400	$0.023	$0.058	$0.116	$0.155
1,000-page PDF	258,000	$0.077	$0.194	$0.387	$1.03
1-hour video, default resolution	about 360,000	$0.108	$0.27	$0.54	$1.44

........

The 1,000-page PDF and the hour of video show how the Pro pricing threshold works: at 258,000 and 360,000 tokens the prompt passes 200,000 tokens, so the whole input is billed at the $4.00 rate instead of $2.00. The same document costs about 19 cents on Gemini 3.8 Flash during December and about 39 cents after the January price change. One hour of video at high resolution comes to about 1,080,000 tokens at the documented rate of 300 tokens per second, close to the size of a full 1M-token context window, so recordings longer than an hour fit only at the default resolution. Full rate cards are in our Gemini API pricing guide, and the context limits per model are covered in our guide to Google AI Studio context window and token limits.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁤

HOW TO UPLOAD AND MANAGE FILES FOR GOOGLE AI STUDIO.

A short checklist avoids most rejected uploads. Start by checking the file against the right limit: 50 MB and 1,000 pages for a PDF, 2 GB for anything going through the Files API, 15 MB for an external URL. Then decide on the route. Files under the inline ceiling that are used once can travel inside the request, while anything larger or reused across prompts should be uploaded through the Files API and referenced by its file URI.

The second step is planning for expiry. A file uploaded on Monday morning is gone by Wednesday morning, so scheduled jobs either upload again at each run or register a Cloud Storage URI, which is fetched per request and can stay registered for up to 30 days. Listing the project's files and deleting the ones no longer needed keeps the project under the 20 GB storage cap.

The third step is format choice. Converting Word files and slide decks to PDF keeps their tables and figures visible to the model, while plain-text formats such as TXT, Markdown and HTML lose their layout. For long video, lowering the frame rate or using the default low resolution keeps a recording within the context window, and splitting a PDF longer than 1,000 pages into parts keeps each part inside the per-file ceiling.

··········

GOOGLE AI STUDIO FILE UPLOAD AND READING LIMITS IN NOVEMBER 2026.

The verified limits draw a clear line between the developer and consumer products. Google AI Studio and the Gemini API accept files of up to 2 GB through a free Files API with 20 GB of storage per project and a fixed 48-hour retention, plus inline requests of up to 100 MB, Cloud Storage registration and 15 MB external URLs. PDFs stop at 50 MB or 1,000 pages at 258 tokens per page, audio at 9.5 hours per prompt, and video at about 3 hours at low resolution on models with a 1M-token context window. The Gemini app stays at 10 files per prompt and 100 MB per file, with video and audio length tied to the Google AI Pro and Ultra plans.

Two points remain open. Google's own pages disagree on the inline size limit for images, 100 MB on the file input methods page against 20 MB on the image understanding page, and the Help Center does not publish the size of the rolling upload limits in the Gemini app. Costs, rather than upload limits, are where the next change is already scheduled: the input price of Gemini 3.8 Flash doubles from $0.75 to $1.50 per 1M tokens on January 1, 2027, which doubles the cost of reading the same file on that model.

··········

FOLLOW US FOR MORE.

·····

DATA STUDIOS

·····

datastudios.org