AWS adds an Australian access path for GPT-5.6 Sol, with a routing trade-off

Amazon Bedrock now lets applications in Sydney and Melbourne invoke GPT-5.6 Sol through `global.openai.gpt-5.6-sol`, its global cross-Region inference profile. Terra and Luna receive analogous profiles. An application makes its request to the Australian Bedrock Runtime endpoint, then Bedrock routes processing to a supported commercial AWS Region. AWS describes Sol as suited to demanding reasoning, coding, and agentic work, while it positions Terra for balanced production use and Luna for high-volume, latency-sensitive workloads.

Australia is the latest documented access route in a wider Bedrock rollout. AWS's August guide says Sol, Terra, and Luna support cross-Region inference in more than 25 AWS Regions, and that an inference profile defines both a model and the Regions AWS may use. The profile lets a caller keep one source Region while AWS selects destination compute from a larger pool. That separates an application's deployment location from the capacity used to process an individual request.

The integration changes the client shape more than the model taxonomy. AWS says Bedrock Runtime accepts Responses, Chat Completions, and Converse calls; an existing OpenAI SDK client can target its OpenAI-compatible route and pass an inference-profile ID as `model`. OpenAI's model documentation separately identifies `gpt-5.6` as an alias routed to Sol and lists a 1,050,000-token context window. Teams should therefore test at the provider boundary: code that assumes a raw OpenAI model name must use the Bedrock profile named in its deployment.

Capacity is not a free compliance abstraction. AWS says a global profile can route to supported commercial Regions based on real-time capacity and that data processed through it may cross the eligible Region set. It recommends a geographic profile or direct single-Region call where data-residency rules constrain processing, and it warns that profile membership and model availability can change. A production rollout should include profile discovery and IAM or service-control-policy checks, then measure logs, latency, and cost by profile rather than treating a Sydney or Melbourne endpoint as a guarantee of processing locality.