Invoking the proprietary OpenAI GPT-5.x models on the bedrock-mantle endpoint returns 401 access_denied, while the open-weight models work from the same account, credentials, endpoint and region. I have worked through the official troubleshooting article and every customer-side prerequisite checks out, so I would like to confirm whether anything is left on my side.
THE ERROR - POST https://bedrock-mantle.us-east-1.api.aws/openai/v1/responses, model openai.gpt-5.6-terra:
401 access_denied - "openai.gpt-5.6-terra is not available for this account. You can explore other available models on Amazon Bedrock. For additional access options, contact AWS Sales."
Request ID: d7ce9714-0de6-48ae-9ac1-4530c230041f
Identical for openai.gpt-5.5, and identical with SigV4 instead of a Bedrock API key.
THE CONTROL - same key, same endpoint, same region, same minute:
openai.gpt-oss-120b -> HTTP 200
openai.gpt-5.6-terra -> 401 access_denied
openai.gpt-5.5 -> 401 access_denied
So this is not IAM, not the bearer token, not the endpoint, and not the base path.
VERIFIED ON MY SIDE:
- Model offered in region: openai.gpt-5.6-terra IS listed in GET /v1/models for us-east-1 (also us-west-2, us-east-2; absent in all EU regions).
- Correct base path: /openai/v1/... for GPT-5.x, /v1/... for gpt-oss.
- Service Quotas: applied value EQUALS the AWS default (20M input TPM / 2M output TPM for GPT-5.6 Terra). Not a quota limit, and nothing to increase.
- Marketplace agreement accepted; GetFoundationModelAvailability returns AVAILABLE / AUTHORIZED / entitlement AVAILABLE.
- License Manager (us-east-1) shows "OpenAI GPT-5.6 Terra (Amazon Bedrock Edition)" as AVAILABLE.
- AWSServiceRoleForAWSLicenseManagerRole was missing and has been created. This made the license visible but did NOT change the 401.
- Use-case form submitted via PutUseCaseForModelAccess.
The troubleshooting article says that when the region is correct, the account has not yet been granted access or capacity, that this is decided at the account level and re-evaluated over time, and that it should be requested through the account team.
MY QUESTIONS:
- Is there anything left on the customer side that I have missed, or is this purely an account-level entitlement?
- Does building usage on the models that ARE available (e.g. openai.gpt-oss-*) actually influence this evaluation, or is that unrelated?
- What is the working path for an account on Basic Support with no assigned account team? I submitted the AWS Sales contact form on 2026-08-24 (no callback) and opened an Account-and-Billing case on 2026-08-25 (no response after 18 hours). Neither channel appears to reach a team that can act on this.
Context: a Turkish-language voice assistant for an online tutoring service. The model is the text layer of a speech-to-text -> LLM -> text-to-speech pipeline, supervised, low volume.
