GPT-5.6 reaches Kiro, while Bedrock availability remains variant- and region-specific

OpenAI says the GPT-5.6 family, including Sol, Terra, and Luna, is available in Kiro, a software-development agent. Its product note describes Kiro as a workflow that turns requirements into technical designs and executable tasks, while retaining review checkpoints and property-based testing. The concrete change is therefore integration into a structured coding environment, rather than a new claim about the underlying capabilities of any one variant. For teams already working in Kiro, the cited material establishes that the named GPT-5.6 variants can be used across planning, implementation, review, and testing work.

The timeline also shows why availability should be read at the product and deployment level, not as one uniform condition. OpenAI's August 24 account covers Kiro and identifies a Kiro-specific Terminal-Bench 2.1 result for Terra. AWS's August 27 technical post then documents India geographic inference profiles for Terra and Luna, while describing global profiles in India that include Sol, Terra, and Luna. The local India profiles are named `in.openai.gpt-5.6-terra` and `in.openai.gpt-5.6-luna`, so a team with Indian data-residency requirements cannot treat family-level naming as a substitute for checking the applicable inference profile.

The comparison worth preserving is one of scope. OpenAI reports a roughly 82% cost reduction for GPT-5.6 Terra on Terminal-Bench 2.1 in Kiro, and attributes the workflow to Kiro's requirements, design, and task context. The cited announcement does not present that figure as a general price reduction for Sol or Luna, nor as a direct Sol-versus-GPT-5.5 benchmark. That distinction matters because an integration result can be useful evidence for a Kiro evaluation without establishing portable performance or cost expectations for direct API use, another agent environment, or a different model variant.

AWS provides the more operational part of the picture. Its India profiles keep inference routing between Mumbai and Hyderabad, support a one-million-token context window with text and image input and text output, and work through OpenAI-compatible Responses and Chat Completions APIs as well as Bedrock Converse. AWS also states that the default zero-data-retention model has an abuse-detection exception for flagged content. For technical users, the practical implication is that Kiro adoption and Bedrock deployment answer different questions: the former concerns a structured development workflow, while the latter requires a variant, region, routing, logging, retention, and API-compatibility review before a production migration.