AWS adds India-bound GPT-5.6 inference profiles for Terra and Luna

Amazon Bedrock now offers India geographic cross-Region inference profiles for GPT-5.6 Terra and Luna. The profile IDs, `in.openai.gpt-5.6-terra` and `in.openai.gpt-5.6-luna`, can be called from AWS's Mumbai or Hyderabad Regions. AWS says the profiles use the Bedrock Runtime endpoint and support OpenAI Responses and Chat Completions APIs alongside Bedrock's Converse API, so an existing application can retain a familiar request shape while changing its endpoint and model identifier.

The important operational change is the routing boundary. AWS describes the India profiles as capacity pooling across Mumbai and Hyderabad while keeping prompts and outputs within India, rather than pinning work to one Region. That is useful for teams that need more throughput without sending inference to a global pool, but it is not single-Region processing: data may move between the two India Regions. AWS also says that its zero-data-retention model has an abuse-detection exception for flagged GPT-5.6 content, which should be considered in a data-handling review.

This is narrower than AWS's global GPT-5.6 profiles, which can route requests to supported commercial Regions worldwide and include Sol as well as Terra and Luna. The India profiles cover Terra and Luna only, making the choice a trade between a geographic processing constraint and the broader capacity or model selection of global routing. AWS's Terra documentation also distinguishes the Runtime path, where geographic inference IDs are used, from the separate Mantle endpoint, so a deployment cannot assume that changing only an AWS Region will enable the new routing behavior.

For technical users, the practical benefit is that one Bedrock integration can combine an OpenAI-compatible client, AWS authentication and India-only cross-Region routing. The limitation is configuration and governance work: IAM must permit the inference profile and the underlying model in every destination Region, and an organization that blocks Regions through service control policies must allow both Mumbai and Hyderabad. Teams should therefore treat the profiles as a residency-oriented architecture option, validate their policy scope and logging expectations, and avoid equating an India geography profile with a guarantee that processing remains in one chosen availability zone or Region.