AWS announced that customers in the Canada (Central) and Canada West (Calgary) regions can now use OpenAI’s newest GPT models through Amazon Bedrock. The feature, called cross‑Region inference (CRIS), routes calls from a Canadian source region to destination regions in the US where the models run. This gives Canadian developers access to GPT‑5.6 (Sol, Terra, Luna), GPT‑6 (Astra, Sol, Luna) and GPT‑6.1 Sol without leaving the AWS network.
How it works and what’s available
You send a request to the Bedrock Runtime endpoint in your Canadian region. Bedrock then forwards it to a supported US region based on the inference profile you choose. Two profile types exist: a US‑geographic profile (keeps traffic within North America) and a global profile (draws from any commercial AWS region). All listed models support text and image inputs, a 1‑million‑token context window, reasoning mode, and tool calling. They also support prompt caching to lower cost for repeated prompts.
Pricing follows the standard Amazon Bedrock rates. Permissions are granted via IAM policies such as the managed AmazonBedrockLimitedAccess policy or custom least‑privilege policies. Organizations using Service Control Policies can allow CRIS traffic while still restricting direct calls to other regions.
Getting started
To try the models, you need an AWS account with ca‑central‑1 or ca‑west‑1 enabled, an IAM role with bedrock:InvokeModel permissions, and Python 3.9+ with the openai, boto3 and aws‑bedrock‑token‑generator packages. Example code shows how to call the OpenAI Responses API, Chat Completions API, or Bedrock Converse API by pointing the client at the regional endpoint and supplying the inference‑profile ID (for example, “global.openai.gpt‑6‑astra”).
Why it matters
Canadian developers can now build generative‑AI apps with the latest OpenAI models without waiting for a local deployment, keeping latency low and staying within North‑American data‑residency rules. They still need to manage IAM permissions and may need to request higher token‑per‑minute quotas for production workloads.
