Amazon Bedrock users in two Canadian AWS Regions can now invoke OpenAI GPT-5.6, GPT-6 and GPT-6.1 through cross-Region inference.
On October 20, 2026, AWS announced that customers using Amazon Bedrock from Canada (Central) and Canada West (Calgary) can access OpenAI’s GPT-5.6, GPT-6 and GPT-6.1 model families through cross-Region inference. The available options include GPT-5.6 Sol, Terra and Luna; GPT-6 Astra, Sol and Luna; and GPT-6.1 Sol.
Customers send requests through a Bedrock endpoint in their Canadian source Region. AWS routes each request to a supported destination Region, drawing on a wider pool of inference capacity without requiring applications to manage the destination routing. Users can select a US geographic profile, which limits processing to Canada and the United States, or a global profile that can route to supported commercial AWS Regions. Profile availability varies: for example, the global profile for GPT-5.6 is available from Canada (Central), while Calgary users must use its US geographic profile.
The models can be called through the OpenAI Responses API, Chat Completions API or Amazon Bedrock Converse API. AWS says they accept text and image inputs, support a one-million-token context window and provide reasoning and tool-calling features. Prompt caching is also supported; eligible prefixes of at least 1,024 tokens can be cached automatically, with subsequent cached-token reads billed at a discount.
For Canadian customers, data at rest—including invocation logs, knowledge bases and configurations—remains in the source Region by default, while inference computation occurs in the destination Region. AWS says requests are encrypted in transit, and model providers do not ordinarily have access to prompts or completions. One exception concerns content flagged by automated abuse classifiers, which may be retained in the destination Region for up to 30 days. Teams with stricter geographic requirements can choose the US profile and should account for destination-region permissions and model-specific quotas when configuring deployments.