AWS Opens GPT-5.6 Access on Amazon Bedrock From Australian Regions

Amazon Web Services said on September 2, 2026, that teams in Australia can now access OpenAI’s GPT-5.6 models on Amazon Bedrock, invoking the Sol, Terra, and Luna variants from the Asia Pacific (Sydney) and Asia Pacific (Melbourne) Regions through global cross-Region inference.

Under the arrangement, an application calls the Amazon Bedrock Runtime endpoint in Sydney or Melbourne, and Bedrock routes the request to a supported commercial AWS Region for processing. AWS said this gives Australian customers access to a broader capacity pool without requiring applications to manage destination Region routing. Three global inference profiles cover the models: global.openai.gpt-5.6-sol, global.openai.gpt-5.6-terra, and global.openai.gpt-5.6-luna. Sydney carries the Region code ap-southeast-2 and Melbourne ap-southeast-4.

The Three GPT-5.6 Variants

AWS described the three variants as serving different workload profiles. According to the AWS Machine Learning Blog post, GPT-5.6 Sol is suited to demanding reasoning, coding, and agentic workloads; Terra balances performance and cost for everyday production use; and Luna provides fast, affordable inference for high-volume and latency-sensitive applications. All three accept text and image inputs, generate text, and support context windows of up to 1 million tokens.

From the two Australian Regions, developers can invoke the models through three access paths on the Bedrock Runtime endpoint: the OpenAI Responses API, the OpenAI Chat Completions API, and the Amazon Bedrock Converse API. The OpenAI-compatible APIs are called on the endpoint’s /openai/v1 paths rather than through AWS SDKs, and the endpoint accepts either AWS Signature Version 4 signing or an Amazon Bedrock model inference API key.

Prompt caching is available for GPT-5.6 through the supported APIs in two modes. Implicit caching is enabled by default with no code changes, while explicit caching lets developers define the reusable prefix, cache boundary, and cache key. AWS noted that profile membership and model availability can change, and it directed customers to its cross-Region inference support documentation to verify configurations before deployment.

Codex Integration and OIDC Authentication

OpenAI’s Codex coding agent can use the same global inference profiles through the Bedrock Runtime model provider built into the latest Codex CLI. AWS said it validated the configuration with codex-cli 0.149.1 running GPT-5.6 Sol from Sydney.

For organizations that federate identity through Okta, Auth0, Microsoft Entra ID, Amazon Cognito, or AWS IAM Identity Center, AWS provides a sample credential helper that exchanges an OpenID Connect token for temporary AWS credentials. Codex then reads those credentials through the standard AWS credential chain, and requests are signed with SigV4, so no API key is involved in the inference path. When the profile is backed by IAM Identity Center, the credentials are already short-term and rotate with the single sign-on session.

Prerequisites for Australian deployments include an AWS account with Sydney or Melbourne enabled as the source Region, an IAM role or user with permissions to invoke the GPT-5.6 inference profiles, and Python 3.9 or later with the openai, boto3, and aws-bedrock-token-generator packages installed. Organizations using service control policies need to verify that their policy allows the GPT-5.6 global inference profiles in the selected source Region. Administrators can confirm active profiles through the AWS CLI or the Amazon Bedrock console’s inference profiles view.

Quotas, Monitoring, and Logging

GPT-5.6 on-demand quotas are measured in requests per minute and tokens per minute, with token burndown determining how each request consumes the token quota. For GPT-5.6, input tokens and cache-write input tokens count at a one-to-one rate, while each output token consumes 10 tokens from the quota, according to AWS. Quotas are reviewed and increased through the Service Quotas console in the source Region the application uses, and AWS advised customers to request increases early, monitor utilization, and test representative prompts, streaming behavior, concurrency, and peak traffic before production rollout.

Because GPT-5.6 requests use the Bedrock Runtime API, calls made through the global inference profiles appear in model invocation logging like other on-demand requests, with records including the inference profile ID and invocation metadata. Codex exports metrics over the OpenTelemetry protocol, and CloudWatch Coding Agent Insights provides a dashboard for that telemetry, covering token usage, API requests, active users, conversation activity, and cache hit rate.

AWS offers two configuration paths for the dashboard: a bearer-token approach using a CloudWatch metrics API key, and an enterprise rollout in which a local collector signs the export with SigV4 using the developer’s federated credentials. AWS classifies the metrics API key as a long-term credential and recommends it only where short-term credentials are not feasible. The enterprise path is the recommended option for organizations that federate developer identity through corporate single sign-on, AWS said.