The available models are GPT-5.6 Sol, Terra and Luna, GPT-6 Astra, Sol and Luna, and GPT-6.1 Sol. With cross-Region inference, an application calls the Bedrock Runtime endpoint in its source Region, and Bedrock routes the request to a supported destination Region. AWS presents this as a way to draw on a larger capacity pool during demand spikes, without the application managing routing itself.
Developers choose between a US geographic inference profile, which keeps processing within Canada and the US, and a global profile, which uses commercial Regions where the model is deployed. The GPT-5.6 global profile is available only from ca-central-1, so Calgary users must use the US geographic profile for those models. The models support a 1-million-token context window, reasoning mode and tool calling.
Quotas are measured in tokens per minute per model, and output tokens count at a burndown rate, so output-heavy workloads use quota faster. Invocation logs and CloudWatch metrics stay in the source Region, while CloudTrail records the destination Region. AWS notes that content flagged by abuse classifiers may be retained for up to 30 days in the destination Region.
Why it matters if you build agents
Agent developers in Canada can now use OpenAI-compatible clients against a regional Bedrock endpoint for these GPT models. Teams with data-residency requirements should pick the US geographic profile and review the 30-day retention note for flagged content. Request quota increases before production rollout and account for output-token burndown when sizing workloads.
Get stories like this every weekday
One short email with the day's agent tooling news.