I'm setting up Amazon Bedrock for a company integration, but every Converse or InvokeModel request to Claude Haiku 4.5 fails immediately with a 429 error about daily token limits. In Service Quotas, the issue appears to be that "Cross-region model inference requests per minute for Anthropic Claude Haiku 4.5" has an applied account-level value of 0, while the default AWS quota is 10,000. The account is about 15 days old, Anthropic model access is already marked as granted, and the problem seems to be the quota rather than model approval. I opened a support case about a week ago but haven't received a useful response. Is an applied quota of 0 normal for newer accounts, and what is the most effective way to escalate a quota issue when the integration is blocked?
4 Answers
Unfortunately, several people have reported long delays getting Bedrock quotas corrected on new accounts. If this is blocking production work, using Anthropic’s API directly may be the fastest workaround. The pricing can be comparable, and it avoids waiting for the Bedrock account quota to be provisioned.
If you have a business or enterprise support relationship, involve your TAM or account team and clearly mark the case as a production-blocking service-quota issue. Without that level of support, escalation may be slow, so documenting the failed calls and requesting confirmation of why the applied quota is zero is probably the best path through the normal support channel.
A zero applied quota can happen with newer accounts, but it effectively prevents the model from being used until AWS finishes whatever account-level review or provisioning is required. Quota-increase requests can generally be submitted even without an enterprise plan, although basic support may provide limited help with service issues. Keep the case updated with the exact quota name, region, account age, model ID, and the fact that every request returns 429.
Double-check that the Anthropic model access request has actually been completed. Some Anthropic models have required a separate approval step beyond enabling Bedrock access. If the console already shows “Access granted,” though, then this probably isn’t an access-approval problem and support will need to address the quota assignment.
Access is already granted for the Anthropic models in the console, so the issue appears limited to the quota being applied as zero.
I had a similar experience with Bedrock quota support and eventually switched to the direct Anthropic API. It was much faster, and I didn’t see a meaningful cost difference.