OpenAI Brings GPT-5.6 Sol, Terra and Luna to Kiro for Agentic Development

GPT-5.6 arrives in Kiro

OpenAI announced on August 24, 2026 that the GPT-5.6 family is now available in Kiro, AWS's software-development agent. The integration includes GPT-5.6 Sol, Terra and Luna, giving developers multiple capability and cost profiles inside a workflow built around requirements, technical design, implementation tasks, review and testing.

This is a developer-platform availability update rather than a new GPT-5.6 model generation. OpenAI originally released GPT-5.6 in July; the Kiro update changes where and how teams can use the family for long-running software work.

Why Kiro's structure matters

Kiro converts high-level product intent into explicit requirements, technical designs and executable tasks. OpenAI says that structured context helps GPT-5.6 stay grounded in what a team is building, the constraints of the codebase and the expected implementation.

That matters for agentic coding because model quality is only one part of the system. Repository context, specifications, task decomposition, tool access, review loops and tests can materially affect whether a coding agent completes useful work without drifting from the original requirements.

Sol, Terra and Luna for different stages

The three GPT-5.6 variants are intended to let teams trade off capability, speed and cost across the software-development lifecycle. A difficult architecture or debugging task may justify a more capable tier, while routine edits, test generation or repetitive implementation work may fit a lower-cost option.

OpenAI frames the integration around price-performance rather than a single benchmark winner. For production teams, the useful comparison is likely to be cost per accepted change or completed engineering task, not token price in isolation.

Reported Terminal-Bench economics

OpenAI says AWS evaluated GPT-5.6 Terra in Kiro on Terminal-Bench 2.1 and found an approximately 82% reduction in cost for successful tasks compared with the reference point described in the announcement. That figure is a vendor-reported result from a particular Kiro evaluation setup and should not be assumed to transfer directly to every repository or engineering workflow.

Teams should measure end-to-end outcomes including retries, human review, test failures, latency and the amount of generated code that is actually accepted. Long-running agent tasks can make these operational costs as important as raw model pricing.

Release status

GPT-5.6 Sol, Terra and Luna are released and available in Kiro according to OpenAI's August 24 announcement. Developers can use the models within Kiro's structured agentic-development workflow today where the service is available.