ChatGPT file upload limits in 2026: size, count and file types

OpenAI sets different caps by file type, plan and where you attach the file. A failed upload usually hits one of the smaller limits long before 512 MB. Every number below was checked against the OpenAI Help Center in October 2026.

The 512 MB ceiling applies to documents and presentations, but most files hit a format-specific cap first. Text and document files such as PDF, DOCX and TXT stop at 2,000,000 tokens, so a text-heavy file far below 512 MB can still be too long.

Spreadsheets skip the token cap. CSV, TSV, XLS and XLSX files top out at roughly 50 MB instead, and wide rows reach that sooner. Images have their own cap of 20 MB each.

For data analysis, ChatGPT also reads JSON, XML, YAML and Markdown files. Parquet is not on OpenAI's list, so convert .parquet files to CSV or JSON before uploading.

Small files run into count limits instead. OpenAI gives no single number of files per conversation; it varies by plan, model and upload type. These caps are fixed, though OpenAI may lower the rate limits during peak hours:

• Rolling rate: up to 80 files every 3 hours. Failed attempts can count toward it, and ChatGPT does not show how much of it is left.
• Free plan: 3 file uploads per day.
• Storage: 25 GB per user and 100 GB per organization. Check your usage in Settings > Storage.
• Projects: 5 files on Free, 25 on Go and Plus, 40 on Pro, Business, Enterprise and Edu.
• Custom GPTs: up to 20 knowledge files, with the same size caps.

For data analysis, ChatGPT writes and runs Python code in a Jupyter notebook environment that can read the files you attached. That environment cannot make web requests, so it cannot download a file from a link you paste, whether the file sits on S3, GitHub, Google Drive or rows.page.

OpenAI also notes that a file can upload fine and still be too large or complex for a complete analysis. If an answer looks partial, ask ChatGPT to work on specific columns or rows, or split the file into smaller ones.

rows.page runs DuckDB-WASM in your browser tab, so the file stays on your machine unless you click Share. It opens CSV, TSV, JSON, JSONL and Parquet, with up to 1.5 GB per tab for the text formats; Parquet is not held to that check. It does not open Excel files, so save a big workbook as CSV first.

1. Drop your .csv, .tsv, .json, .jsonl or .parquet file onto rows.page.
2. Check the Columns panel for types, empty values and top values.
3. Filter rows in the filter box, for example country:NL or price:>20, or click a histogram bar.
4. Or type a DuckDB SQL query that starts with SELECT, WITH or FROM. The table is called data.
5. Open Export and pick CSV, JSON, JSONL or Parquet. Only the current filtered view or SQL result is exported.

This query collapses a large event log into one row per day, country and status:

SELECT
  date_trunc('day', created_at) AS day,
  country,
  status,
  count(*) AS events,
  sum(amount) AS total_amount,
  median(amount) AS median_amount
FROM data
WHERE created_at >= '2026-01-01'
GROUP BY ALL
ORDER BY day DESC, events DESC

If ChatGPT needs raw rows, drop wide text columns and keep a subset. Exclude the hidden __rid row number too, or DISTINCT treats every row as unique:

SELECT DISTINCT * EXCLUDE (__rid, raw_payload, debug_log)
FROM data
WHERE status = 'error'
LIMIT 50000

A filtered view exports every matching row, while a SQL result exports up to 1,000,000 rows. CSV exports are comma-delimited UTF-8 without a BOM.

If you only upload to ChatGPT to show someone a slice of the data, share it from rows.page instead. Click Share in the explorer, or run curl -T data.csv https://rows.page, to get a link like https://rows.page/<10-character id> for files up to 100 MB.

Without an account you get 50 pushes and 1 GB per IP address per UTC day, and links expire after 7 days. The filter, sort and visible columns live in the URL, so Copy link on the dataset page shares exactly what you see. Sign in with Google to keep a dataset on your account, or pick Pro ($12 per month) or Max ($39 per month) for private datasets.