View a markdown version of this page

Costs and throughput - Elemental Inference

Costs and throughput

A function that invokes a model calls Amazon Bedrock in your account, with your credentials. That means:

  • The model charges appear on your own Amazon Bedrock bill. AWS CloudTrail in your account records each call under the feed's access role, in a role session named ei-function-feed-id.

  • Elemental Inference invokes the model in the feed's AWS Region. The invocations count against your account's Amazon Bedrock quotas for that model in that Region. To use a model in other Regions, specify a cross-Region inference profile ID as the function's model.

  • Your account must be able to invoke the model in that Region before the function can run. Elemental Inference doesn't enable model access on your behalf.

Important

On-demand Amazon Bedrock throughput is shared across everything in your account that uses that model. A function on a live feed fires continuously for as long as the channel runs, which is the workload that exhausts an on-demand quota and starts throttling your other applications. For a production feed, use provisioned throughput for the model, or a cross-Region inference profile, and size it for the cadence you configured. For more information, see Increase throughput with provisioned throughput in the Amazon Bedrock User Guide.

A passthrough function invokes no model, so it incurs no model charges and consumes no Amazon Bedrock quota.

For more information about Elemental Inference pricing, see Pricing for AWS Elemental Inference.