Skip to content

Claude Opus 5 returns “not available for this account” despite AVAILABLE agreement, entitlement, authorization and region

0

We are trying to use Claude Opus 5 on Amazon Bedrock from Asia Pacific (Sydney), ap-southeast-2.

The Australian inference profile is visible and active:

au.anthropic.claude-opus-5

We also tested:

global.anthropic.claude-opus-5

Both runtime invocations fail with:

AccessDeniedException: anthropic.claude-opus-5 is not available for this account. You can explore other available models on Amazon Bedrock. For additional access options, contact AWS Sales.

However, the Bedrock control-plane access checks now show that all required availability conditions are satisfied.

Running:

aws bedrock get-foundation-model-availability
--region ap-southeast-2
--model-id anthropic.claude-opus-5

returns:

{ "modelId": "anthropic.claude-opus-5", "agreementAvailability": { "status": "AVAILABLE" }, "authorizationStatus": "AUTHORIZED", "entitlementAvailability": "AVAILABLE", "regionAvailability": "AVAILABLE" }

We also successfully created the foundation-model agreement using:

aws bedrock create-foundation-model-agreement
--region ap-southeast-2
--model-id anthropic.claude-opus-5
--offer-token "<offer-token>"

The command completed successfully and returned:

{ "modelId": "anthropic.claude-opus-5" }

The agreement initially showed PENDING and later changed to AVAILABLE.

We then tested actual invocation through Bedrock Runtime using the Converse API.

Example:

import boto3

client = boto3.client( "bedrock-runtime", region_name="ap-southeast-2" )

response = client.converse( modelId="au.anthropic.claude-opus-5", messages=[ { "role": "user", "content": [ {"text": "Reply exactly OK"} ] } ], inferenceConfig={ "maxTokens": 8, "temperature": 0 } )

This still returns:

AccessDeniedException: anthropic.claude-opus-5 is not available for this account.

We also tested the global profile:

global.anthropic.claude-opus-5

and received the same account-level error.

For comparison, the same AWS account, environment and Bedrock Runtime setup successfully invokes:

au.anthropic.claude-opus-4-6-v1

with a normal successful response.

We also tested newer Claude models and observed the following:

Model Agreement Authorization Region Runtime invoke Claude Opus 4.6 AVAILABLE AUTHORIZED AVAILABLE PASS Claude Opus 4.7 NOT_AVAILABLE AUTHORIZED AVAILABLE FAIL Claude Opus 4.8 NOT_AVAILABLE AUTHORIZED AVAILABLE FAIL Claude Sonnet 5 AVAILABLE AUTHORIZED AVAILABLE FAIL Claude Opus 5 AVAILABLE AUTHORIZED AVAILABLE FAIL

For Sonnet 5, all relevant control-plane fields also showed available/authorized, but runtime invocation still returned:

anthropic.claude-sonnet-5 is not available for this account

This suggests there may be an account-level runtime access restriction that is not represented by get-foundation-model-availability.

Questions Is Claude Opus 5 currently subject to a staged rollout, usage-history requirement, runtime allowlist, capacity gate, or other account-level eligibility restriction on Amazon Bedrock? Can get-foundation-model-availability return: agreement = AVAILABLE authorization = AUTHORIZED entitlement = AVAILABLE region = AVAILABLE

while the account is still excluded from runtime model access?

If so, how can a customer on AWS Basic Support request enablement for: au.anthropic.claude-opus-5 Is there any additional Marketplace, Anthropic, Bedrock, billing or account prerequisite that is not exposed through get-foundation-model-availability? Is there a recommended escalation path for Basic Support customers when AWS Sales cannot investigate account-level Bedrock model access?

We are specifically interested in Australian processing through the AU geographic inference profile, but the same failure also occurs with the Global Opus 5 profile.

Any clarification from AWS or other customers who have encountered the same situation would be appreciated.

2 Answers
0

The delivery infrastructure (endpoint) changed starting with Opus 4.7.

Anthropic's official documentation is split into two pages based on this transition. The current page covers Claude on Bedrock provided via the Messages API (/anthropic/v1/messages), while the legacy Bedrock integration (using InvokeModel or Converse APIs with ARN-format model IDs) has been moved to a separate document for "Opus 4.6 and earlier" versions.

Delivery PathInvocation Method
Opus 4.6 and earlierbedrock-runtimeConverse / InvokeModel
Opus 4.7 and later (4.7 / 4.8 / Sonnet 5 / Opus 5 / Fable 5)bedrock-mantleMessages API

You are using boto3.client("bedrock-runtime").converse(), which corresponds to the legacy path. This fully explains why Opus 4.6 works while versions 4.7 and later are rejected. The seemingly contradictory situation where Sonnet 5 failed despite having open access permissions is also due to this same reason (the difference in endpoints).

answered 5 days ago

0

Thanks — the Anthropic documentation split is a useful clue. However, AWS’s current Opus 5 documentation explicitly lists bedrock-runtime as supported and indicates that both Converse and InvokeModel are supported, including the au.anthropic.claude-opus-5 inference profile. AWS’s endpoint documentation also continues to support bedrock-runtime for these invocation patterns.

Anthropic’s Bedrock documentation likewise indicates that newer Claude models, including Opus 4.7, 4.8, Opus 5 and Sonnet 5, remain accessible through Bedrock Runtime APIs.

So the Messages/Mantle transition does not appear, by itself, to explain the account-level not available for this account response.

We will test the bedrock-mantle Messages endpoint as an additional diagnostic. If Mantle succeeds while bedrock-runtime fails, that would point to an endpoint-specific restriction. If Mantle returns the same account-level error, that would provide stronger evidence of an account/runtime eligibility restriction.

One additional clue is that Sonnet 5 also fails for this account despite its control-plane availability and authorization checks succeeding.

Thanks again — the endpoint distinction is definitely worth testing.

answered 4 days ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.