We don't publish
your competitive advantage.

IFWhen using LangChain's OpenAICallbackHandler with OpenAI o1 or o3-mini reasoning models, reasoning tokens are excluded from completion token cost calculation, leading to underestimated costs.

THENIn the token cost calculation logic (typically in openai_info.py), retrieve `reasoning_tokens` from `usage_metadata.output_token_details` and add them to `completion_tokens` before computing `completion_cost`. Ensure the model name is standardized and the total cost reflects both regular completion tokens and reasoning tokens.

IFWhen using LiteLLM proxy server, streaming calls do not report cost on the client side (cost is missing from headers or final chunk).

THENSet the client-side configuration flag `litellm.include_cost_in_streaming_usage = True` before making streaming requests. This ensures the cost is included in the last chunk of the streaming response, similar to OpenRouter's behavior.

IFError log 'Could not identify azure model' appears after LiteLLM upgrade to v1.75.5-stable for Azure OpenAI calls.

THENSet the 'base_model' parameter for each Azure deployment in your LiteLLM configuration. This resolves the error and ensures accurate token/cost tracking. The correct documentation link is https://docs.litellm.ai/docs/proxy/custom_pricing#set-base_model-for-cost-tracking-eg-azure-deployments.

IFWhen using LiteLLM proxy server with enterprise license, streaming calls do not return cost information on the client side.

THENSet the flag `litellm.include_cost_in_streaming_usage = True` in the client SDK to enable cost tracking for streaming responses. This ensures the final chunk includes usage cost, similar to OpenRouter.