Sorting LiteLLM GitHub issues by difficulty to implement. Higher = harder.

{
The left issue is harder because it requires provider-specific request/response translation across differing Gemini API semantics, preservation of reasoning metadata, and regression coverage for multiple model generations. The right issue is comparatively localized: define and raise a typed exception at the unmapped-model path while maintaining compatibility with existing callers.
}
https://github.com/berriai/litellm/issues/32769 4:1 https://github.com/berriai/litellm/issues/29146