ChatGPT Upload Failed? New File Size Limits and Supported 2026 Formats
If you just saw the “Upload Failed” or “Unknown Error” message while attaching a document, you aren’t alone. As we move further into 2026, OpenAI has updated the infrastructure for GPT-5.3, introducing a complex “Dual-Limit” system: one for Megabytes and one for Tokens.
Understanding these boundaries is the only way to avoid the “red bar of death” during your research or data analysis.
Quick Summary of 2026 Limits:
- Max File Size: 512MB (Docs/Code).
- Max Image Size: 20MB.
- Token Cap: 2,000,000 tokens per file (the “Invisible Limit”).
- Storage Cap: 10GB per user (Total history).
- Rate Limit: 80 files / 3 hours (Plus tier).
Why Your Upload Actually Failed: The 3 Main Reasons
1. The “Token Ceiling” (The 2M Rule)
This is the #1 cause of failures in 2026. You might have a 100MB PDF that contains a million words. Even though the file is small in megabytes, it exceeds the 2 million token processing limit. Fix: Split the PDF into chapters or use an OCR tool to simplify the text before uploading.
2. Forbidden Formats
While ChatGPT supports almost everything, some formats still trigger errors:
- Folders: You cannot upload a folder. You must ZIP it first (yes, ZIP is fully supported in 2026!).
- Proprietary Formats: Files like
.pages,.epub, or password-protected PDFs will fail immediately. - Corrupted Meta-data: If a file was partially downloaded, ChatGPT’s 2026 security filter will reject it.
3. The 20MB Image Wall
If you are trying to upload high-resolution RAW photos or massive 4K screenshots for analysis, they will fail. All images—regardless of your plan—are capped at 20MB per file.
Supported Formats in March 2026
ChatGPT has expanded its library. Here is what is officially supported today:
| Category | Formats | Notes |
|---|---|---|
| Documents | PDF, DOCX, TXT, RTF | Best for summaries and Q&A. |
| Data | CSV, XLSX, JSON, XML | Capped at ~50MB for stable analysis. |
| Code | PY, JS, CPP, HTML, CSS… | Fully searchable and executable in Canvas. |
| Images | PNG, JPG, WEBP, GIF | Multimodal analysis only (no SVG support). |
| Archives | ZIP, TAR | ChatGPT will “unzip” and read the contents. |
Pro-Tips to Avoid Upload Errors
1. The “Flattening” Technique
If you have a complex Excel file with 20 tabs, it will likely crash. Flatten it into a single CSV. GPT-5.3 processes CSV files 3x faster and with 90% fewer errors than multi-tab XLSX files.
2. Use “Cloud Connect” (Google Drive/OneDrive)
In 2026, the Google Drive integration is often more stable than direct local uploads. If a file fails from your desktop, try importing it directly from the cloud icon. This uses a server-to-server transfer which bypasses browser-side timeouts.
3. Clear Your “AI Storage”
Go to Settings > Data Controls > Connected Files. If you have reached your 10GB limit, new uploads will fail without a clear explanation. Delete old Project assets to make room for new ones.
Conclusion: Smaller is Better
Even though GPT-5.3 can “read” more than any previous model, the web interface is still the bottleneck. For the most stable experience, keep your PDFs under 25MB and your datasets under 50MB.
Are you getting an error specifically with long PDFs? Check out our guide on How to Handle 100k+ Token Prompts to learn how to chunk your data for the AI!