fix: enable metadata parameter for OpenAI API calls

[emerzon]

Enable metadata parameter for OpenAI API calls

Fixes issue where metadata was not being sent to OpenAI API despite being passed to litellm.completion(). The metadata field is now properly supported for OpenAI and tracked in logging integrations like Langfuse.

Relevant issues

Fixes #12997

Type

🐛 Bug Fix

Changes

• Add metadata to OPENAI_CHAT_COMPLETION_PARAMS and DEFAULT_CHAT_COMPLETION_PARAM_VALUES
• Add metadata to OpenAIGPTConfig.get_supported_openai_params()
• Add metadata parameter to get_optional_params() function signature
• Pass metadata through to optional_params in completion flow
• Remove preview features gate for metadata support
• Ensure metadata is available in logging for all integrations

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
litellm	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jul 25, 2025 11:25pm

[krrishdholakia]

Hey @emerzon this was intentional - as the metadata param is also used by litellm for internal logic. We are trying to migrate away from this to litellm_metadata. Do you have a use-case today blocked because of this?

[emerzon]

Hi @krrishdholakia, yes, we're trying to setup Session tracking in Langfuse, and it does require this feature: https://langfuse.com/docs/observability/features/sessions

[delacroixe]

We are trying to send metadata to LangFuse to add more information. This change would be great to achieve our goal.

[hnt2601]

Hey @emerzon this was intentional - as the metadata param is also used by litellm for internal logic. We are trying to migrate away from this to litellm_metadata. Do you have a use-case today blocked because of this?

same

[pbebbo]

Hey @emerzon, do you know when this bug will be fixed?

[chrisk314]

Hi, I'm wondering what's the status of this issue? We execute workflows consisting of hundreds of LLM requests in a B2B app. We use metadata to efficiently filter the traces in Arize Phoenix. This works well using LlamaIndex but with LiteLLM we lose the metadata making it impractical to work with the traces.

[emerzon]

Hi @krrishdholakia, any visibility on this issue?

[pbebbo]

@krrishdholakia and @emerzon any updates on this?

[emerzon]

Closing in favor of #16523