OpenAI GPT models now reachable on Amazon Bedrock from Canadian AWS Regions

Which models are available

AWS lists three families of OpenAI models reachable from Canada: GPT-5.6 (Sol, Terra and Luna), GPT-6 (Astra, Sol and Luna) and GPT-6.1 Sol. The company describes each by positioning. GPT-6.1 Sol is aimed at strong reasoning and coding where performance and cost both matter. GPT-6 Astra targets complex end-to-end reasoning and agentic tasks, while GPT-6 Luna is meant for fast, low-cost inference on high-volume repeated tasks, with adjustable reasoning effort per request.

In the GPT-5.6 line, Sol is positioned for high reasoning, coding and agentic workloads, Terra for balanced performance and cost in everyday production, and Luna for fast and affordable use in high-volume, latency-sensitive applications.

According to AWS, the models accept text and image inputs, return text, and support a 1-million-token context window, reasoning mode and tool calling. GPT-6.1 Sol has the same context window and tool calling, plus additional input and output modalities documented in its model card. Prompt caching is also supported on Amazon Bedrock for all three families.

How cross-Region inference works

AWS describes cross-Region inference (CRIS) as a capacity mechanism. An application calls the Amazon Bedrock Runtime endpoint in its source Region, and Amazon Bedrock routes the request to a supported destination Region for processing. This gives access to a broader capacity pool that can help sustain throughput during demand spikes, and developers do not need to handle destination Region routing in their own code.

Two inference profile types are offered from Canada. The US geographic profile, prefixed us., routes requests to the Canadian source Region and US Regions. AWS recommends it when a workload requires inference processing to stay within North America. The global profile, prefixed global., routes to supported commercial AWS Regions where the model is deployed.

Availability differs by model. GPT-6.1 Sol and the GPT-6 models are served by both profile types from ca-central-1 and ca-west-1. The global profile for the GPT-5.6 models is available only from ca-central-1, so users in ca-west-1 need the US geographic profile for GPT-5.6. AWS notes that profile membership and model availability can change and advises checking active profiles before deployment, either with the AWS CLI or on the console.

Setup, APIs and access control

The models can be called through the OpenAI Responses API, the OpenAI Chat Completions API and the Amazon Bedrock Converse API on the bedrock-runtime endpoint. The OpenAI-compatible APIs use the /openai/v1 paths. The endpoint accepts either AWS Signature Version 4 or an Amazon Bedrock API key, and the post's examples use a token generator for Python to create a short-term key from existing AWS credentials, so no static key has to be stored.

Prerequisites include an AWS account with one of the two Canadian Regions enabled as the source Region, an IAM role or user allowed to invoke the inference profiles, and Python 3.9 or later with the openai, boto3 and aws-bedrock-token-generator packages. For quick exploration, AWS suggests the managed policy AmazonBedrockLimitedAccess, and a least-privilege custom policy for deployed workloads.

Organizations that restrict Regions with service control policies can use the bedrock:InferenceProfileArn condition key to allow CRIS traffic while blocking direct calls to unapproved Regions. AWS states that every destination Region in the chosen profile must be allowed for a request to succeed.

What it means for Canadian teams

For Canadian organizations building on Amazon Bedrock, the change lets them use these OpenAI models from Canadian Regions, with the US geographic profile available for workloads that must keep processing within North America. The post also says it covers streaming and tool-use examples for the Responses API, quota increase requests and observability in the source Region. Pricing and per-model Region details are referred to the Amazon Bedrock pricing page and the model cards in the Amazon Bedrock User Guide.