Accessing OpenAI models on Amazon Bedrock from Australia with global cross-Region inference
Amazon Web Services Frank Huang ● Covered by 2 sources
OpenAI’s latest models are now available on Amazon Bedrock in Sydney and Melbourne. AWS is routing the traffic for you, so Australian teams get more capacity without juggling Regions.
Based on reporting by Amazon Web Services, Frank Huang — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Australian teams can now reach OpenAI’s newest models through Amazon Bedrock without leaving local AWS Regions. The setup works from Asia Pacific (Sydney) and Asia Pacific (Melbourne), while Amazon Bedrock sends the request on to a supported commercial AWS Region behind the scenes.
AWS is offering three models here: GPT-5.6 Sol, Terra, and Luna. Sol is aimed at heavier reasoning, coding, and agent-style work. Terra is pitched as the middle ground for everyday production use. Luna is the quick and cheaper option for high-volume jobs where latency matters. All three take text and image inputs, return text, and support context windows of up to 1 million tokens.
The practical win is less about the model names and more about how they’re exposed. Applications can call the Bedrock Runtime endpoint in Sydney or Melbourne and use the Responses API, Chat Completions API, or Converse API. The OpenAI-compatible paths live under /openai/v1, and the endpoint accepts either SigV4 or a Bedrock model inference API key. AWS also points to its token generator so developers can create a short-lived key instead of parking a static one in code.
AWS is leaning hard into operational details too. The post walks through prompt caching, Codex setup with OIDC-based authentication, and monitoring through CloudWatch and Coding Agent Insights. It also spells out quota mechanics: GPT-5.6 on-demand limits are tracked in requests per minute and tokens per minute, with output tokens consuming 10 tokens from quota. Before production, AWS wants teams testing concurrency, streaming, prompt length, and peak traffic, which is usually where the nice demo stops being nice.
There’s a catch, of course. Profile membership and model availability can change, so AWS tells users to check cross-Region inference support before deploying. But for Australian teams already standardised on Bedrock, this is a neat way to get OpenAI models without building their own routing layer.
My take — AI-written commentary, not fact-checked reporting
This is what cloud vendors do best when they’re behaving themselves: hide the messy plumbing and charge for the convenience. The bigger story isn’t the model list, it’s that AWS is turning access, routing, auth, and telemetry into one tidy control plane. That’s catnip for enterprises, and a reminder that “model choice” now looks a lot like choosing your landlord.
Read more about this at: Amazon Web Services