What shipped
OpenAI and AWS made the full GPT-5.6 family available inside Kiro on August 24. Developers can choose among Sol, Terra and Luna while working through Kiro's requirements, technical-design and task-planning workflow.
The product positioning matters because it places model selection inside a software-delivery system. Kiro supplies repository context and team standards, then creates checkpoints where a developer can review plans and implementation rather than accepting one large unstructured patch.
The 82% cost claim needs context
OpenAI reports that GPT-5.6 Terra completed successful Terminal-Bench 2.1 tasks in Kiro at roughly 82% lower cost. The announcement attributes the result to fewer missteps and better grounding from Kiro's spec-driven context.
That figure should be read as a measured result under the partners' test configuration. It does not establish the same saving for every language, repository or task. Cost depends on token volume, retries, tools, test execution and the amount of human correction required before a patch is accepted.
Why specifications can change agent performance
Coding agents often lose time because the requested behavior, constraints and definition of done remain implicit. A structured specification can reduce that ambiguity before the model edits files, while an executable task list gives reviewers smaller units to inspect.
Structure can also add overhead. A tiny one-file change may not need a formal design document. Teams should reserve the full workflow for changes where unclear requirements, multiple components or long execution make early planning cheaper than later rework.
What engineering teams should test
Build a fixed evaluation set from recent completed work: a bug fix, a feature spanning several files, a test-writing task and a maintenance change. Compare accepted completion rate, wall-clock time, model cost, test pass rate and reviewer correction time against the current tool.
Record the chosen GPT-5.6 variant and settings. Sol, Terra and Luna trade capability, speed and cost differently, so a statement about the family is not enough to support a procurement decision. The relevant result is cost per merged, correct change in the team's own environment.
Sources & further reading
Social-media activity is treated as a signal of attention, not proof. Product claims are attributed to the linked publisher or announcement.