AWS opens Canadian Bedrock access to OpenAI GPT models

AWS has made OpenAI GPT-5.6, GPT-6 and GPT-6.1 models accessible to Amazon Bedrock customers from Canada Central and Canada West through cross-Region inference.

Using a Canadian Bedrock endpoint does not mean inference stays in Canada. The change lets applications draw on model capacity outside the source region, with the choice of a US-geographic or global inference profile setting the routing boundary.

The available variants are GPT-5.6 Sol, Terra and Luna; GPT-6 Astra, Sol and Luna; and GPT-6.1 Sol. The release continues the expansion of Bedrock’s managed model catalog, following AWS’s addition of GLM 5.3 to Amazon Bedrock.

Applications send requests to the Bedrock Runtime endpoint in Canada Central (ca-central-1) or Canada West in Calgary (ca-west-1) and include an inference-profile identifier. Bedrock then chooses a destination region included in that profile; the application does not select the processing region itself.

A us.-prefixed geographic profile can route requests to the Canadian source region and US regions. AWS describes that boundary as North American processing, not Canada-only processing. A global.-prefixed profile can route requests among supported commercial AWS regions worldwide, offering a broader capacity pool without a geographic processing restriction.

The options also vary by Canadian source region. Both Canada Central and Canada West can use the US geographic profile for GPT-5.6, but the global profile is available only from Canada Central. A GPT-5.6 workload originating in Canada West must therefore use the US profile.

For deployments with residency requirements, AWS instructs customers to call GetInferenceProfile from every source region they plan to use and inspect the regions in the returned model ARNs. AWS says the destination membership of a geography-bound profile does not change, although it may introduce a new profile with different membership. A global profile’s destination set can expand as AWS adds commercial regions.

AWS says invocation logs, knowledge bases and configurations remain at rest in the Canadian source region by default, while transient model inference occurs in the destination region selected by the profile. Cross-Region traffic stays on the AWS network and is encrypted in transit. CloudTrail records remain in the source region and identify the processing region in the additionalEventData.inferenceRegion field.

The covered OpenAI models have a separate retention consideration. AWS says inputs and outputs flagged by its classifiers may be retained for up to 30 days for automated offline abuse detection. With cross-Region inference, that retained material is stored in the destination region where the request was processed.

AWS documentation says global cross-Region inference costs about 10% less for input and output tokens than geographic cross-Region inference, with charges based on the source region.