What Changed
Amazon Bedrock now offers Australian teams access to three OpenAI GPT-5.6 models — Sol, Terra, and Luna — through global cross-region inference profiles available from both the Asia Pacific (Sydney) and Asia Pacific (Melbourne) AWS regions. Applications call the Amazon Bedrock Runtime endpoint in either Australian region, and Bedrock handles routing to a supported commercial AWS region for processing, removing the need for applications to manage destination-region logic themselves.
All three models accept text and image inputs, produce text output, and support context windows of up to one million tokens. According to AWS, Sol targets demanding reasoning, coding, and agentic workloads; Terra is positioned for everyday production use balancing performance and cost; and Luna is designed for high-volume, latency-sensitive applications where speed and affordability matter most.
How Invocation Works
Developers can reach these models through three API paths on the Amazon Bedrock Runtime endpoint: the OpenAI Responses API, the OpenAI Chat Completions API, and the Amazon Bedrock Converse API. The OpenAI-compatible paths are served under /openai/v1 and accept either AWS Signature Version 4 or a short-term Amazon Bedrock model inference API key — the latter can be generated dynamically from existing AWS credentials using the aws-bedrock-token-generator Python package, avoiding the need to store static keys.
Prompt caching is also available. Implicit caching is on by default with no code changes required, while explicit caching lets developers define reusable prefixes, cache boundaries, and cache keys to further manage inference costs.
Quota and Monitoring Considerations
On-demand quotas are measured in requests per minute and tokens per minute. Notably, output tokens carry a higher quota cost than input tokens — each output token consumes ten tokens from the TPM quota under GPT-5.6's burndown rate. AWS recommends requesting quota increases before production rollout and testing representative workloads including streaming, concurrency, and peak traffic patterns.
Because requests use the standard Amazon Bedrock Runtime API, they appear in model invocation logging alongside other on-demand calls. For teams using OpenAI Codex, CloudWatch Coding Agent Insights provides a dashboard covering token usage, API requests, active users, cache hit rates, and optional organizational dimensions such as team, department, and cost center.
Who This Affects
This availability is most directly relevant to Australian organizations already using or evaluating OpenAI models that want to keep data flows within AWS infrastructure and leverage existing IAM, SCP, and CloudWatch tooling. Teams using Codex for coding assistance can configure it to use the Amazon Bedrock Runtime provider and route through Sydney or Melbourne with OIDC-based federated authentication — supported identity providers listed by AWS include Okta, Auth0, Microsoft Entra ID, Amazon Cognito, and AWS IAM Identity Center.
One practical consideration for smaller teams: the quota burndown model means output-heavy workloads — such as long code generation tasks — may exhaust TPM limits faster than input-heavy ones, so monitoring and early quota planning could be important before scaling.
Profile membership and model availability can change, so AWS advises checking cross-region inference support documentation before deployment.
Read the official announcement (opens in a new tab)