What are the image limits for ChatGPT, Claude, Gemini, Copilot and Perplexity?
From each provider's help center or documentation, October 2026
| Assistant | Max per image or file | Other limits |
|---|---|---|
| ChatGPT | 20 MB per image | 512 MB hard limit per file, all file types |
| Claude (claude.ai) | 10 MB per image | 20 images per message; 8,000 × 8,000 px maximum |
| Claude API | 10 MB per image (base64-encoded); 5 MB on Amazon Bedrock and Google Cloud | Up to 600 images per request (100 on 200k-context models); 2,000 × 2,000 px maximum when a request has more than 20 images; 32 MB per request |
| Gemini app | 100 MB per file (videos up to 2 GB) | Up to 10 files per prompt |
| Microsoft Copilot | 50 MB per file | Up to 20 files per conversation; images as PNG or JPEG |
| Perplexity | 40 MB per file | Weekly upload allowances depend on the plan |
Note the Claude API detail: the 10 MB counts the base64-encoded image, which is about a third bigger than the file, so the file itself should stay under about 7.5 MB. Claude accepts JPEG, PNG, GIF and WebP; animated GIFs are read from their first frame only.
Why does a smaller image work just as well?
Because the model downscales big images before it looks at them. Anthropic's documentation states the rule for Claude: images beyond 1,568 px on the long edge (about 1,568 visual tokens) are scaled down on standard-resolution models, and beyond 2,576 px (4,784 visual tokens) on the high-resolution models, Claude 4.7 and later. Each 28 × 28 pixel patch costs one token, so the cost grows with resolution up to that cap:
Claude visual tokens per image on high-resolution models (Anthropic vision documentation)
| Image size | Processed at | Visual tokens |
|---|---|---|
| 1,000 × 1,000 px | Not resized | 1,296 |
| 1,920 × 1,080 px | Not resized | 2,691 |
| 2,000 × 1,500 px | Not resized | 3,888 |
| 3,840 × 2,160 px | 2,576 × 1,449 px | 4,784 |
A 12-megapixel phone photo is several megabytes of upload that the model shrinks anyway. On the APIs, larger images also cost more input tokens up to the cap, and every resent image in a long conversation adds to the request size.
What size should I send?
- Photos: 1,500–2,000 px on the long edge, JPEG or WebP at quality 80. That is usually 200–600 KB.
- Screenshots with text: keep at least 1,500 px of width for a full screen, and quality 85–90 or PNG. Anthropic warns that heavy JPEG compression can make text hard to read.
- Documents, receipts, whiteboards: crop to the page first, so the pixels go to the text rather than the table around it.
How do I shrink a screenshot or photo before uploading?
- Open the compressor in your browser. On iPhone, use Safari so HEIC photos open.
- Drop the screenshot or photo.
- Turn on Limit width and type 2000. For a portrait phone photo, whose long side is the height, type 1500 to get 2,000 px tall.
- Choose WebP or JPG as the output and set Quality to 80, or 85–90 for screenshots full of text.
- Compress, download and upload the new file. If the app's limit is the problem rather than tokens, Image to 2 MB gets any photo under 2 MB in one step.
Why does the upload still fail when the file is small enough?
- Format: HEIC photos and some TIFF or AVIF files are not accepted everywhere. Save as JPG or WebP.
- Too many images: claude.ai takes 20 per message; Gemini 10 files per prompt; Copilot 20 files per conversation.
- Plan allowances: free plans limit how many files you can upload per day or week.
- Dense screenshots: split a very tall page capture into two images rather than shrinking it until the text blurs; the screenshot compression guide has settings for text.