Toolofficial publisher

code_execution (Gemini)

Tool : A precise action the AI can call · requires gemini-api

Python executed on Google's side, results returned to the model.

Practical and included; isolated execution at Google.

Docs

Overall score

7.0 /10

Security: one caveat, see the card

Utility

8.0

Reliability

6.0

Security

One caveat

Ease

1 click · Comfortable

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

When a language model needs to perform precise calculations, process files, or solve equations, text generation alone is not enough. The code_execution tool in the Gemini API allows the model to generate Python code, run it in an isolated environment hosted by Google, and then use the results to iteratively refine its response.

In practice, when you enable the tool in your request, Gemini can write a Python script, send it to a secure runtime environment, retrieve the output, and continue reasoning from it. The model can repeat this cycle up to five times if an error occurs. Execution is limited to 30 seconds per call.

This tool is aimed at developers integrating the Gemini API who need reliable calculations, CSV or text file processing, or Matplotlib chart generation directly within the model's response. It also works in multi-turn conversations through the previous_interaction_id parameter.

However, it is not suitable for those who want to execute code in a language other than Python: Gemini can generate code in other languages, but cannot run it through this tool. It is also not intended for end users without API access.

What it offers

·Python code execution in an isolated environment hosted by Google, at no additional cost beyond the tokens consumed.
·Iterative reasoning: the model can re-run execution up to five times to correct errors before providing a final response.
·Support for file input (CSV, text) and Matplotlib chart generation returned as inline images in the response.
·Image processing with Gemini 3 Flash: the model can write and execute Python code to crop, annotate, or analyze images passed as input.
·Use in multi-turn conversations via the previous_interaction_id parameter to chain exchanges with context memory.
·Compatibility with Python, JavaScript SDKs and the REST API for flexible integration depending on the development environment.
·Maximum execution time of 30 seconds per call, with automatic error return if this limit is exceeded.

Our tips

For calculations on CSV or text files, pass the file as inline data or via the Files API: files too large for the model's context window will trigger an error.
Explicitly ask the model to use code for tasks such as counting elements or rotating an image: automatic zoom on small details is the only behavior triggered without an explicit instruction.
In a multi-turn conversation, pass the previous_interaction_id from the previous exchange so the model retains context and can chain calculations.
Matplotlib charts generated are returned as inline images in the response: plan in your client code to handle the image content type to display or save them.
The tool can only execute Python: if you need another language, you can request code generation, but execution will have to be done client-side.

Watch out

Only Python can be executed through this tool: any code in another language generated by the model cannot be run in the execution environment.
Execution is capped at 30 seconds and input files are limited by the model's context window, which can block processing of large files.

In practice

Price: Freemium
Commercial use: Commercial use: yes
Available on: api
Host: gemini-api
FR interface: No

Security and data

Rating: Security: one caveat, see the card
Capabilities: readswritesexecutesnetwork
Your data: Data with the vendor
Maintenance: Active

Links and repository checked on 25 October 2026 · added on 1 June 2024

How to install code_execution (Gemini)

1. Obtain a Gemini API key from Google AI Studio and store it in an environment variable (GEMINI_API_KEY) rather than hard-coding it. 2. Install the Python SDK with: pip install google-genai 3. In your script, import the client: from google import genai 4. Instantiate the client: client = genai.Client() 5. Add the tool to your request by passing tools=[{"type": "code_execution"}] when calling client.interactions.create(). 6. Iterate over interaction.steps to separately retrieve the text, generated code, and execution results.

pip install google-genai

Official documentation ↗

In the same category

Whole category Feed (memory, context) →