Dynamic output_type when using FallbackModel

[marmor7]

Description

I really prefer using NativeOutput over ToolOutput when possible.

So basically whenever we're using an OpenAI model we set output_type = NativeOutput(...)
However, we want a fallback model that is from a different provider, assuming that the fallback will be used when the primary provider (e.g. OpenAI) servers are currently down, so no use in falling back on a different model from the same provider.

But when i set FallbackModel("openai:...", "anthropic:...")
i can no longer use NativeOutput.

This is related to #3104 but that one is about a bug where two models that support NativeOutput are erroneously returning a not-supported error.

My feature request is to allow me to set NativeOutput when the primary model is in use, and ToolOutput when the fallback model is in-use.

Thanks!

References

No response

[DouweM]

@marmor7 Thanks for the report, I'm looking into this and #3104 now.