OutYet reporting
GPT-5.6 Sol gains an Australian Bedrock route, with global routing caveats
AWS has documented a Bedrock path for GPT-5.6 Sol from Sydney and Melbourne, but the global profile can process requests outside those source Regions.
AWS published a September 2 guide describing access to GPT-5.6 Sol through Amazon Bedrock from the Asia Pacific (Sydney) and Asia Pacific (Melbourne) Regions. The guide names the global inference profile `global.openai.gpt-5.6-sol`, alongside Terra and Luna profiles. This is a regional Bedrock-access update, not evidence from OutYet's detector that establishes or changes the model's release state.
The regional entry point should not be confused with an in-country processing guarantee. AWS says an application sends its request to the Bedrock Runtime endpoint in Sydney or Melbourne, then Bedrock routes it to a supported commercial AWS Region. Its cross-Region inference documentation likewise says a global profile automatically selects a commercial Region for processing, so teams with data-residency requirements need to inspect the routing model rather than infer residency from the endpoint they call.
For teams already using OpenAI-shaped clients, AWS documents three invocation paths: the OpenAI Responses API, OpenAI Chat Completions API, and Bedrock's Converse API. The OpenAI-compatible routes use Bedrock Runtime's `/openai/v1` endpoint and can authenticate with SigV4 or a Bedrock model inference API key. That gives existing applications a migration path, but it also makes AWS identity, permissions, and endpoint configuration part of the integration.
The guide also gives Codex users a concrete deployment pattern: configure Codex with the Bedrock Runtime provider, point it at `global.openai.gpt-5.6-sol`, and use an AWS profile backed by an OIDC credential process. AWS says the helper exchanges an identity-provider token for temporary AWS credentials, after which requests are signed with SigV4. That pattern may suit organizations that want inference access governed through AWS federation instead of distributing a static inference credential.
Capacity and quota behavior remain operational constraints. AWS says GPT-5.6 on-demand quotas are measured in requests per minute and tokens per minute, with output tokens consuming quota at a higher burndown rate than input tokens. It also warns that inference-profile membership and model availability can change, and recommends checking support, requesting quota increases early, and testing representative prompts, output lengths, streaming, concurrency, and peak traffic before production rollout.