Enable Code Execution in the API

Run Python code via the Grok API

Code Execution lets the model write and run Python code in real time within a secure sandboxed environment. Instead of simply generating code for you to run yourself, the model executes the code directly and returns the results. It’s a powerful tool for calculations, data analysis, and code validation.

Tool identifiers by SDK

The tool identifier varies depending on the SDK you use:

This naming difference matters. Use the identifier that matches your SDK to avoid errors.

Basic activation

cURL / Direct JSON

{
  "model": "grok-4.20-0309-reasoning",
  "input": "Calculate the sum of the first 1,000 prime numbers",
  "tools": [
    {"type": "code_interpreter"}
  ]
}

xAI SDK (Python)

response = client.responses.create(
    model="grok-4.20-0309-reasoning",
    input="Calculate the sum of the first 1,000 prime numbers",
    tools=[client.tools.code_execution()]
)

print(response.output_text)

OpenAI SDK (Python)

response = client.responses.create(
    model="grok-4.20-0309-reasoning",
    input="Calculate the sum of the first 1,000 prime numbers",
    tools=[{"type": "code_interpreter"}]
)

Vercel AI SDK (JavaScript)

const response = await generateText({
  model: xai("grok-4.20-0309-reasoning"),
  prompt: "Calculate the sum of the first 1,000 prime numbers",
  tools: [xai.tools.codeExecution()]
});

How the model uses Code Execution

When Code Execution is enabled, the process works as follows:

1. The model analyzes the request and determines that a calculation or data analysis is needed
2. It writes Python code suited to the problem
3. The code runs in xAI’s Python sandbox
4. The model receives the execution results (standard output, any errors)
5. It formulates its response incorporating the results

If the code produces an error, the model can automatically correct and re-execute it—which may generate multiple invocations for a single request.

Temperature recommendation

For mathematical calculations and data analysis, xAI recommends using low temperature values (0.0 to 0.3). A low temperature reduces variability in code generation and produces more reproducible results.

response = client.responses.create(
    model="grok-4.20-0309-reasoning",
    input="Calculate the Sharpe ratio for this portfolio...",
    tools=[client.tools.code_execution()],
    temperature=0.1
)

Key takeaways

• Code Execution enables real-time Python execution in a secure sandbox
• The identifier varies by SDK: code_execution (xAI), code_interpreter (OpenAI)
• The model writes, executes, and interprets the code automatically
• Use grok-4.20-0309-reasoning and a low temperature for calculations
• The cost is $0.005 per execution, with multiple invocations possible