[Feature] A simple working example of GPT-5 with Azure?

[Gby56]

What feature would you like to see?

Hi ! I have been trying desperately to get azure openai with GPT-5 working, but still no luck.
The model name/api base is highly confusing, plus the max_completion_tokens vs max_token is also not helping

If anyone got it working with LiteLLM under the hood, please help !

Would you like to contribute?

• Yes, I'd like to help implement this.
• No, I just want to request it.

Additional Context

No response

[Gby56]

here's my working hack for azure

def _patch_dspy_kwargs_for_reasoning_models() -> None:
    """Patch DSPy to handle kwargs properly for reasoning models.

    Ensures that max_tokens is not passed in kwargs for reasoning models like gpt-5.
    Instead, max_completion_tokens should be used.
    """
    original_init = LM.__init__

    def patched_init(
        self: 'LM',
        model: str,
        model_type: str = 'chat',
        temperature: float = 0.0,
        max_tokens: int = 4000,
        cache: bool = True,
        is_reasoning_model: bool = True,
        **kwargs: Any,
    ) -> None:
        if is_reasoning_model:
            # Remove any max_tokens from kwargs to prevent conflicts
            kwargs.pop('max_tokens', None)

        # Call original init first
        result = original_init(
            self, model, model_type=model_type, temperature=temperature, max_tokens=max_tokens, cache=cache, **kwargs
        )

        # Now override the kwargs that DSPy set incorrectly
        if is_reasoning_model:
            # For reasoning models, ensure we use max_completion_tokens and remove max_tokens from kwargs
            self.kwargs = dict(temperature=temperature, **kwargs)
            # Explicitly remove max_tokens if it was added anywhere
            self.kwargs.pop('max_tokens', None)
        else:
            # For non-reasoning models, use max_tokens
            self.kwargs = dict(temperature=temperature, max_tokens=max_tokens, **kwargs)

        return result

    # Apply the monkey patch
    LM.__init__ = patched_init

[TomeHirata]

Have you checked LiteLLM's doc for Azure OpenAI? https://docs.litellm.ai/docs/providers/azure/#overview
What's the exact issue you're facing?