Code Interpreter / Advanced Data Analysis: What It Can And Can't Do

Code Interpreter (also known as Advanced Data Analysis) is a Python sandbox running on OpenAI's servers. The model writes code, executes it, sees the output, and iterates. You get the equivalent of a junior data analyst with read-only access to the files you upload. The model cannot reach the open internet from inside the sandbox.

Things it does brilliantly

• Cleaning and exploring CSV / Excel data — pivot tables, joins, summary stats.
• Generating charts from a dataset and iterating on them ('make the bars sorted, add labels').
• Converting between file formats — PDF to text, Excel to CSV, image to text via OCR libraries.
• Running statistical tests when you describe the question in plain English.
• Writing and testing one-off Python scripts before you copy them into your own environment.

Things it cannot do

• Reach the public internet from inside the sandbox — no API calls, no scraping.
• Persist files across separate conversations — the sandbox is reset.
• Run very long jobs — there are wall-clock timeouts in the tens of seconds for any single execution.
• Install arbitrary packages — only what is in the pre-installed environment, though the set is large.
• Process huge files — practical limits depend on tier, but assume a few hundred MB ceiling, often less.

Applied exercise

1. 1Find a CSV you have been meaning to look at — anything with at least 5 columns and 1000 rows.
2. 2Upload it and ask: 'What are the three most surprising things in this data?'
3. 3Ask to see the code for the most interesting answer.
4. 4Verify one number by hand. Note where the model's framing of a column was off, if it was.

The big idea: Code Interpreter is a fast scratchpad, not a production runtime. Use it to think with data, then port the logic out.