- Newest
- Most votes
- Most comments
As far as I know, yes, this is a known behavior with newly launched models. The issue stems from the separation between the Control Plane (entitlements) and the Data/Runtime Plane (actual inference quotas).
What is happening:
- Control Plane (AUTHORIZED): The APIs correctly reflect that your account has successfully accepted the EULA and holds the Marketplace entitlement.
- Data Plane (AccessDeniedException): Despite the authorization, the backend Tokens-Per-Minute (TPM) quota for this specific new model has not been provisioned yet (effectively sitting at 0). This discrepancy triggers the "not available" error at runtime, which can easily be mistaken for an IAM or credential issue.
Because this is a backend discrepancy regarding the runtime quota rather than a simple synchronization delay, waiting or re-triggering the agreement will not resolve it.
How to resolve it:
I would strongly recommend opening an AWS Support case (and involving your TAM/Account Manager if you have one). Provide them with your specific region (us-east-1) and ask them to verify and provision the backend TPM quotas for anthropic.claude-opus-4-8. Once the backend quota is explicitly allocated, the runtime will start accepting requests.
Relevant content
- asked 8 months ago
- asked 4 months ago
- asked 2 years ago
