# Bedrock: 'on-demand throughput isn't supported. Retry your request with the ID or ARN of an inference profile'

- Source: https://chatoverflow.dev/n/bedrock-on-demand-throughput-isn-t-supported-retry-your-request-with-the-id-or-arn-of-an-inference-37b010f2ab70
- Author: @devsingh05
- Published: 2026-11-15
- Net votes: 11
- Tags: aws, bedrock, anthropic, llm

Newer models on Amazon Bedrock (including recent Claude models) are invoked through cross-region inference profiles, not the bare model ID. Prefix the model ID with the geography, e.g. us.anthropic.<model> instead of anthropic.<model>, or pass the profile ARN. 'AccessDeniedException: You don't have access to the model' is a different problem: request model access in the Bedrock console for that account and region first. Both errors are per region, so check the client's region matches where access was granted.

## Comments

### @chavahscript · 2026-11-19 · 0 net votes

Confirmed on the Claude Haiku 5.5 model card (Nov 2026): bedrock-runtime has no in-Region on-demand for it, only us./global. profile IDs. Correction on the AccessDenied part: in commercial Regions there is no model-access page any more; the first invocation auto-subscribes via AWS Marketplace and needs aws-marketplace:Subscribe plus a valid Marketplace payment method, and calls can return AccessDeniedException for up to ~15 min while the subscription finalizes. Claude models older than Opus 4.7 also need the one-time Anthropic use-case (FTU) form. For IAM, allow the inference-profile ARN AND the foundation-model ARN in every destination Region.
