Implementation:BerriAI Litellm Check Batch Cost

Attribute	Value
Sources	enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py
Domains	Cost Tracking, Batches, Background Jobs, Enterprise
Last Updated	2026-02-15 16:00 GMT

Overview

CheckBatchCost is a background polling utility that checks the completion status of managed batch jobs and tracks their cost and usage through the LiteLLM logging pipeline.

Description

This class polls the LiteLLM_ManagedObjectTable database table for batch jobs that are in validating, in_progress, or finalizing status. For each active job, it:

1. Decodes the unified object ID to extract the model ID and batch ID.
2. Queries the LLM provider via llm_router.aretrieve_batch to check the current batch status.
3. If the batch is completed and has an output_file_id:
   • Retrieves the output file content via the managed files hook.
   • Parses the file content into a dictionary.
   • Looks up the deployment to determine the LLM provider and model name.
   • Calls calculate_batch_cost_and_usage to compute cost and token usage from the batch output.
   • Creates a LiteLLMLogging object and calls async_success_handler to log the cost through the standard logging pipeline (which updates spend tracking in the database).
4. Marks completed jobs as complete in the database.

The class requires ProxyLogging, PrismaClient, and Router instances.

Usage

Instantiate and call check_batch_cost periodically (e.g., via a background scheduler) to ensure batch job costs are tracked after completion.

Code Reference

Source Location

enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py

Signature

class CheckBatchCost:
    def __init__(
        self,
        proxy_logging_obj: ProxyLogging,
        prisma_client: PrismaClient,
        llm_router: Router,
    ): ...

    async def check_batch_cost(self): ...

Import

from litellm_enterprise.proxy.common_utils.check_batch_cost import CheckBatchCost

I/O Contract

Inputs

Parameter	Type	Description
proxy_logging_obj	ProxyLogging	Proxy logging object used to access registered hooks (e.g., managed_files).
prisma_client	PrismaClient	Database client to query managed object table.
llm_router	Router	LLM router for retrieving batch status and deployment info.

Outputs

Output	Type	Description
Cost logging	Side effect	Batch cost and usage logged through LiteLLMLogging.async_success_handler.
Status update	Side effect	Completed jobs updated to "complete" status in the database.

Usage Examples

from litellm_enterprise.proxy.common_utils.check_batch_cost import CheckBatchCost

checker = CheckBatchCost(
    proxy_logging_obj=proxy_logging_obj,
    prisma_client=prisma_client,
    llm_router=llm_router,
)

# Run as a periodic background task
await checker.check_batch_cost()

Related Pages

• BerriAI_Litellm_Check_Responses_Cost -- Similar polling for responses API cost tracking
• BerriAI_Litellm_Managed_Files -- Managed files hook used to retrieve batch output files