GPT-5.6 gets two AWS routes: Kiro for structured coding and Bedrock for regional deployment

OpenAI says the GPT-5.6 family, including Sol, Terra, and Luna, is now available in Kiro, AWS's software-development agent. The announcement describes Kiro as a system that turns high-level intent into requirements, technical designs, and executable tasks, then carries that structured context into planning, building, review, and testing. The relevant change is access inside that development environment, rather than evidence of a new model-family launch.

OpenAI's performance claim is deliberately narrow. It says testing found GPT-5.6 Terra completed Terminal-Bench 2.1 tasks in Kiro at roughly an 82% cost reduction. The company attributes the result to Kiro's spec-driven context, which supplies requirements, technical designs, and task context before implementation begins. That makes the figure evidence about the Terra-plus-Kiro workflow, not a general ranking of GPT-5.6 against every coding tool or model.

AWS's August 27 Bedrock announcement shows a separate deployment path. Its India geographic profiles expose Terra and Luna, with profile identifiers for each, while the global profile supports Sol, Terra, and Luna. AWS says the India profiles route only between Mumbai and Hyderabad, allowing capacity to be shared inside the country for workloads with local data-processing requirements. Sol is therefore present in Kiro and AWS global inference, but not in the documented India-specific pair.

The Bedrock integration also changes the engineering trade-off. AWS says Terra and Luna accept text and image input, produce text, and offer a one-million-token context window. It documents OpenAI-compatible Responses and Chat Completions APIs alongside Bedrock's Converse API, so existing OpenAI SDK integrations can point to a Bedrock endpoint while teams that prefer AWS authentication can use the native interface. Prompt caching is supported, with AWS stating a 90% discount for cached input reads.

For technical users, Kiro is the more opinionated option when a team wants planning, checkpoints, and property-based testing around coding work. Bedrock is the more infrastructure-oriented option when API compatibility, regional routing, or AWS operational controls are decisive. The 82% number should not be treated as a universal saving, and data-residency evaluations still need care: AWS says content flagged by its automated abuse-detection classifiers can be retained for offline abuse detection.