completion(): your first call

litellm.completion takes a model name and messages in the OpenAI shape and returns the answer in it. mock_response answers without calling any provider.

from litellm import completion

response = completion(
    model="openai/gpt-4o-mini",
    messages=[{"role": "user", "content": "I was charged twice for one order"}],
    mock_response="billing",
)
print(response.choices[0].message.content)
print(response.choices[0].finish_reason)
print(response.usage)

mock_response is documented for tests: LiteLLM builds a normal response around that text instead of sending the request, so no key is needed. Take it out and set OPENAI_API_KEY, and the same call goes to OpenAI.

The answer is where lesson 1 found it for OpenAI, choices[0].message.content, whichever provider answered. finish_reason says why the model stopped. The mock's usage numbers are made-up placeholders, not counts of this text.

Dictionary style works too

response = completion(model="openai/gpt-4o-mini", messages=[{"role": "user", "content": "hi"}], mock_response="Hello!")
print(response["choices"][0]["message"]["content"])
print(type(response).__name__)

The result is a ModelResponse object that also answers square-bracket access, so code written against OpenAI's JSON keeps working.

Async

import asyncio

from litellm import acompletion


async def main():
    response = await acompletion(model="openai/gpt-4o-mini", messages=[{"role": "user", "content": "hi"}], mock_response="Hello from async")
    print(response.choices[0].message.content)


asyncio.run(main())

acompletion is the same call for async code, so many requests can be waiting at once, as in Python for AI.