OpenAI GPT-5.6 models expand to 25+ AWS Regions via Bedrock

Amazon Web Services Amazon.com, Inc. ↘ 2.16% has launched cross-region inference for OpenAI's GPT-5.6 model family on Amazon Bedrock, expanding access to more than 25 AWS Regions worldwide. The announcement, made on August 17 and detailed in a technical blog post published August 20, marks a substantial scaling of the partnership between AWS and OpenAI to deliver frontier AI models through cloud infrastructure.aws.amazon+1

Three Variants, One Endpoint

The rollout covers three general-purpose GPT-5.6 variants — Sol, Terra, and Luna — each designed for different workload profiles. Sol targets frontier reasoning, autonomous coding, and scientific research. Terra offers balanced performance for everyday production tasks at roughly half the cost of Sol. Luna provides fast, affordable inference for high-volume applications.aws.amazon+1

All three variants accept text and image inputs, support a 1-million-token context window, and work with the OpenAI Responses API, Chat Completions API, and Amazon Bedrock Converse API, including streaming. Developers using the OpenAI SDK can point their existing client at Bedrock's OpenAI-compatible endpoint and swap in an inference profile ID as the model parameter.aws.amazon+1

How Cross-Region Inference Works

Cross-region inference operates through two profile types. Geographic profiles, prefixed with a geography code such as `us.`, route requests only within a defined boundary — for example, across US East (N. Virginia), US West (Oregon), and US East (Ohio) — to satisfy data residency requirements. Global profiles, prefixed `global.`, route to any supported commercial Region based on real-time capacity and carry a roughly 10-percent per-token price discount.aws.amazon+1

AWS also introduced India-specific geographic profiles for Terra and Luna, routing requests within Mumbai and Hyderabad to meet local regulatory requirements.aws.amazon

Broader Context

GPT-5.6 first became generally available on Amazon Bedrock in mid-July, initially through a limited set of Regions. Since then, AWS has progressively expanded capabilities, adding 1-million-token context support on August 3, launching Daybreak cybersecurity variants on August 11, and now enabling cross-region inference with global and geographic routing. The service uses AWS's zero-operator-access security model, with all requests authenticated through IAM and logged in CloudTrail.linkedin+3

The expansion comes as OpenAI scales its compute infrastructure in parallel — the company recently confirmed that its first batch of Nvidia ↘ 0.33% Vera Rubin racks is operational and running training workloads for next-generation models.moomoo