Code Execution

Some models and systems on Groq have native support for automatic code execution, allowing them to perform calculations, run code snippets, and solve computational problems in real-time.

Only Python is currently supported for code execution.

Supported Models

Built-in code execution is supported for the following models and systems:

For a comparison between the compound-beta and compound-beta-mini systems and more information regarding extra capabilities, see the Compound Systems page.

Quick Start (Compound)

To use code execution with Groq's Compound systems, change the model parameter to one of the supported models or systems.

import os
from groq import Groq

client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

response = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "Calculate the square root of 101 and show me the Python code you used",
        }
    ],
    model="compound-beta-mini",
)

# Final output
print(response.choices[0].message.content)

# Reasoning + internal tool calls
print(response.choices[0].message.reasoning)

# Code execution tool call
if response.choices[0].message.executed_tools:
    print(response.choices[0].message.executed_tools[0])

And that's it!

When the API is called, it will intelligently decide when to use code execution to best answer the user's query. Code execution is performed on the server side in a secure sandboxed environment, so no additional setup is required on your part.

Final Output

This is the final response from the model, containing the answer based on code execution results. The model combines computational results with explanatory text to provide a comprehensive response. Use this as the primary output for user-facing applications.

import math
print("The square root of 101 is: ")
print(math.sqrt(101))

Reasoning and Internal Tool Calls

This shows the model's internal reasoning process and the Python code it executed to solve the problem. You can inspect this to understand how the model approached the computational task and what code it generated. This is useful for debugging and understanding the model's decision-making process.

Executed Tools Information

This contains the raw executed tools data, including the generated Python code, execution output, and metadata. You can use this to access the exact code that was run and its results programmatically.

{
  "string": "",
  "name": "",
  "index": 0,
  "type": "python",
  "arguments": "{\"code\": \"import math; print(\"The square root of 101 is: \"); print(math.sqrt(101))\"}",
  "output": "The square root of 101 is: \n10.04987562112089\n",
  "search_results": { "results": [] }
}

Quick Start (GPT-OSS)

To use code execution with OpenAI's GPT-OSS models on Groq (20B & 120B), add the code_interpreter tool to your request.

from groq import Groq

client = Groq(api_key="your-api-key-here")

response = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "Calculate the square root of 12345. Output only the final answer.",
        }
    ],
    model="openai/gpt-oss-20b",  # or "openai/gpt-oss-120b"
    tool_choice="required",
    tools=[
        {
            "type": "code_interpreter"
        }
    ],
)

# Final output
print(response.choices[0].message.content)

# Reasoning + internal tool calls
print(response.choices[0].message.reasoning)

# Code execution tool call
print(response.choices[0].message.executed_tools[0])

When the API is called, it will use code execution to best answer the user's query. Code execution is performed on the server side in a secure sandboxed environment, so no additional setup is required on your part.

Final Output

This is the final response from the model, containing the answer based on code execution results. The model combines computational results with explanatory text to provide a comprehensive response.

Reasoning and Internal Tool Calls

This shows the model's internal reasoning process and the Python code it executed to solve the problem. You can inspect this to understand how the model approached the computational task and what code it generated.

Executed Tools Information

This contains the raw executed tools data, including the generated Python code, execution output, and metadata. You can use this to access the exact code that was run and its results programmatically.

{
  name: 'python',
  index: 0,
  type: 'function',
  arguments: 'import math\nmath.sqrt(12345)\n',
  search_results: { results: null },
  code_results: [ { text: '111.1080555135405' } ]
}

How It Works

When you make a request to a model or system that supports code execution, it:

1. Analyzes your query to determine if code execution would be helpful (for compound systems or when tool choice is not set to required)
2. Generates Python code to solve the problem or answer the question
3. Executes the code in a secure sandboxed environment powered by E2B
4. Returns the results along with the code that was executed

Use Cases (Compound)

Mathematical Calculations

Ask the model to perform complex calculations, and it will automatically execute Python code to compute the result.

import os
from groq import Groq

client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

chat_completion = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "Calculate the monthly payment for a $30,000 loan over 5 years at 6% annual interest rate using the standard loan payment formula. Use python code.",
        }
    ],
    model="compound-beta-mini",
)

print(chat_completion.choices[0].message.content)

Code Debugging and Testing

Provide code snippets to check for errors or understand their behavior. The model can execute the code to verify functionality.

import os
from groq import Groq

client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

chat_completion = client.chat.completions.create(
    messages=[
        {
            "role": "user",
            "content": "Will this Python code raise an error? `import numpy as np; a = np.array([1, 2]); b = np.array([3, 4, 5]); print(a + b)`",
        }
    ],
    model="compound-beta-mini",
)

print(chat_completion.choices[0].message.content)

Security and Limitations

• Code execution runs in a secure sandboxed environment with no access to external networks or sensitive data
• Only Python is currently supported for code execution
• The execution environment is ephemeral - each request runs in a fresh, isolated environment
• Code execution has reasonable timeout limits to prevent infinite loops
• No persistent storage between requests

Pricing

Web search is priced at $0.00005 / second of execution.

Please see the Pricing page for more information.

Provider Information

Code execution functionality is powered by E2B, a secure cloud environment for AI code execution. E2B provides isolated, ephemeral sandboxes that allow models to run code safely without access to external networks or sensitive data.