Amazon Bedrock adds OpenAI GPT-5.6 models and cross-region inference

New OpenAI Model Support

Amazon Bedrock has officially expanded its lineup to include the OpenAI GPT-5.6 models. These specific versions are known as Terra and Luna. They are now available on the bedrock-runtime endpoint, which is the primary way developers interact with models in the AWS ecosystem.

The update specifically targets the AWS GovCloud (US) environment. This means that government entities and regulated industries can now access these specific OpenAI models while staying within a secure, isolated cloud infrastructure. It bridges the gap between high-end commercial AI and strict government security requirements.

By adding these models to the bedrock-runtime, Amazon ensures they are treated as first-class citizens in the platform. Developers can use the same tools and workflows they already use for other models. This reduces the friction of switching between different AI providers or model types within a single project.

Expanded API Functionality

The integration of GPT-5.6 comes with broad API support. Users can now utilize the Responses, Converse, and Chat Completions APIs for these models. These are the standard interfaces used to build applications that can generate text, hold conversations, or process complex instructions.

Because these native OpenAI APIs now run on the bedrock-runtime, they inherit the existing security and management framework. This means that any account-level controls a user has already set up for other models will automatically apply to the new OpenAI models. It creates a unified experience for managing different AI technologies.

This consistency is important for large organizations that manage many different models. It ensures that security teams do not have to create new rules for every new model added to the platform. Everything remains centralized under the same administrative umbrella provided by AWS.

Cross-Region Inference

A major technical addition in this update is cross-region inference for the AWS GovCloud (US) regions. This feature allows the system to automatically route requests across both available regions. The goal is to provide users with higher throughput and increased redundancy without requiring manual management of capacity.

In many cloud setups, users have to manually balance their traffic between different locations to avoid hitting limits. Cross-region inference removes this burden by distributing the load automatically. If one region experiences a demand spike, the system can shift the work to the other region to ensure the application stays online.

This specific implementation is designed to help users scale while maintaining US data sovereignty. Data sovereignty refers to the legal requirement that data remains within specific geographic borders. By staying within the GovCloud (US) regions, the system meets these requirements while still providing the broadest possible access to Bedrock capacity.

Logging and Cost Tracking

Transparency in AI spending and usage is a core part of this launch. Because the OpenAI models are integrated into the bedrock-runtime, all usage data is captured automatically. This includes model invocation logging, which can be delivered to Amazon S3 or Amazon CloudWatch Logs for auditing purposes.

The platform also provides specific metrics in Amazon CloudWatch. These metrics cover essential performance data such as invocation counts, token counts, latency, throttles, and errors. This allows engineers to monitor the health of their AI applications in real-time and identify any bottlenecks quickly.

Finally, the costs associated with using the GPT-5.6 models are itemized in the AWS Cost Explorer and the AWS Cost and Usage Report. This level of detail is crucial for businesses that need to attribute specific spending to different models. It makes it easier to calculate the return on investment for different AI projects across a large company.

Availability and Next Steps

Cross-region inference for OpenAI models is now available in every AWS Region where these models are currently offered on Amazon Bedrock. This ensures that customers can benefit from the new scaling features regardless of their primary geographic location.

For those looking to implement these models, Amazon provides specific documentation to get started. Users are encouraged to review the model cards for GPT 5.6 (Terra and Luna) in the Amazon Bedrock User Guide. These cards provide the necessary technical details and limitations for each model.

This expansion signals a continued effort to integrate high-performing third-party models into the AWS ecosystem. By providing both the models and the infrastructure to scale them, AWS is making it easier for complex organizations to deploy AI safely. The focus remains on providing high throughput and reliability for demanding workloads.