If your team is based in Australia and you’ve been waiting to use OpenAI’s latest models without managing cross-Region routing yourself, there’s a new option. As of September 2, 2026, Amazon Bedrock offers OpenAI GPT-5.6 models—Sol, Terra, and Luna—through global cross-Region inference from both Asia Pacific (Sydney) and Asia Pacific (Melbourne). You call the Bedrock Runtime endpoint in your local Region, and Bedrock handles routing to a supported commercial AWS Region behind the scenes. This means you get access to a larger capacity pool without building your own failover logic.
For product builders, the practical question is: how do you actually invoke these models, and what should you watch out for? This post walks through the setup, the three API paths, and a few operational tips like prompt caching and authentication.
What You Get: Three Models, One Endpoint
GPT-5.6 comes in three flavors, each with a global inference profile ID that you use as the model identifier:
- Sol (
global.openai.gpt-5.6-sol): Built for demanding reasoning, coding, and agentic workloads. - Terra (
global.openai.gpt-5.6-terra): Balances performance and cost for everyday production use. - Luna (
global.openai.gpt-5.6-luna): Fast, affordable inference for high-volume, latency-sensitive apps.
All three accept text and image inputs, generate text, and support context windows up to 1 million tokens. You can invoke them from either ap-southeast-2 (Sydney) or ap-southeast-4 (Melbourne) using the Bedrock Runtime endpoint. The destination Region is chosen by AWS, so you don’t need to track which Region is handling your request.
Before you start, check the Cross-Region inference support page because profile membership and model availability can change. Also, make sure your AWS account has the source Region enabled, your service control policy allows the GPT-5.6 profiles, and you have the right IAM permissions.
Three Ways to Call the Models
You can access GPT-5.6 through three APIs on the Bedrock Runtime endpoint. The OpenAI-compatible APIs live under /openai/v1, and you can authenticate with AWS SigV4 or a short-term Bedrock model inference API key. The AWS blog post shows how to generate that key on the fly using the AWS Bedrock Token Generator for Python, so you don’t have to store a static key.
OpenAI Responses API – If you already use the OpenAI SDK, just point the client at the Regional endpoint:
from aws_bedrock_token_generator import provide_token
from openai import OpenAI
region = "ap-southeast-2" # or "ap-southeast-4" for Melbourne
model_id = "global.openai.gpt-5.6-terra"
client = OpenAI(
base_url=f"https://bedrock-runtime.{region}.amazonaws.com/openai/v1",
api_key=provide_token(region=region),
)
response = client.responses.create(
model=model_id,
input="Explain Availability Zones in three bullet points.",
max_output_tokens=300,
)
print(response.output_text)For streaming, set stream=True and iterate over events.
Chat Completions API – If your app uses the older Chat Completions style, that works too. The example in the blog uses max_completion_tokens and reasoning_effort parameters, which are familiar to OpenAI users.
Amazon Bedrock Converse API – When you’re building on AWS SDKs, use boto3 and the Converse API. It resolves credentials through the standard AWS chain, so no extra setup is needed. Streaming is available via converse_stream.
All three examples work from either Sydney or Melbourne—just swap the Region code.
Prompt Caching and Codex Setup
Two operational details stand out for production use.
Prompt caching is available in two modes. Implicit caching is on by default and requires no code changes. Explicit caching lets you define reusable prefixes and cache keys, which can cut costs for repetitive workloads. The blog points to the GPT-5.6 introduction post for examples.
Codex can also use these global inference profiles. The blog validates a setup with codex-cli 0.149.1 and GPT-5.6 Sol from Sydney. For organizations using Okta, Auth0, Microsoft Entra ID, Amazon Cognito, or IAM Identity Center, there’s a sample OIDC auth helper that exchanges an OIDC token for temporary AWS credentials. You configure it in ~/.aws/config with credential_process, then point Codex at the profile in ~/.codex/config.toml. This avoids storing API keys—requests are signed with SigV4.
Quota Management and Final Thoughts
One thing the blog doesn’t detail is quota management—it’s mentioned as a section header but not filled in. That’s a gap you’ll want to investigate on your own, especially if you’re planning high-volume usage.
For Australian teams, this is a meaningful step: you can now use OpenAI models through Bedrock without leaving your local Region for API calls. The tradeoff is that you’re relying on AWS’s routing decisions, so latency and data residency may vary depending on where requests actually land. If that matters for your workload, test with your own traffic patterns.
Start by verifying the inference profiles are active in your Region using the AWS CLI or console, then try the Responses API with a simple prompt. From there, you can layer in caching and authentication as needed.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.