Can I force the execution of multiple distinct functions in parallel using tool_choice?

Below is an explanation with relevant details from the documentation:

Key Points

• Forcing a Single Function
  When using the tool_choice parameter with a structure like:

  JSONCopied
  tool_choice={
    "type": "function",
    "function": { "name": "function1" },
  }

  this explicitly forces the model to only call "function1". This means its output will be constrained to that particular function call.

• Parallel Function Calling with Auto Mode
  By default, if you allow the model to select the function by setting tool_choice to "auto" (or simply not specifying a forced function with no tools) along with parallel_tool_calls: true (the default setting), the model may return multiple function calls—even for different functions—if the context indicates it's needed. The example in the documentation shows three different weather check calls being produced together:

  PYTHONCopied
  tool_calls=[
      chat.completionsMessageToolCall(..., function=Function(arguments='{"city":"New York"}', name='check_weather')),
      chat.completionsMessageToolCall(..., function=Function(arguments='{"city":"London"}', name='check_weather')),
      chat.completionsMessageToolCall(..., function=Function(arguments='{"city":"Tokyo"}', name='check_weather'))
  ]

  However, note that in this case, the functions were the same (i.e., multiple calls to check_weather) even though with different arguments.

• Forcing Two Different Functions in One Request
  As of now, if you specify tool_choice to force a function call, you can only force one specific function per request. There isn’t a built-in way to enforce two distinct functions to be called simultaneously if you use the forced mode (i.e., specifying one function with tool_choice).
  If your use case requires two different functions to be called, you should:

  1. Set tool_choice to auto (or use the default behavior) so the model can decide which functions to call.
  2. Ensure the desired functions are available in your tools configuration.
  3. The model may then generate multiple function calls in parallel (with parallel_tool_calls: true).

Conclusion

You cannot force two different functions simultaneously using a single tool_choice object that only accepts one function name. If you want to leverage parallel function calling for different functions, you should rely on the auto mode where the model determines which functions to call, or structure your conversation such that multiple function calls (possibly with different function names) are generated as needed.