GPT-5.6 Sol enters Kiro's agentic engineering workflow

OpenAI says the GPT-5.6 family, including Sol, Terra, and Luna, is now available in Kiro. Kiro's model documentation independently lists GPT-5.6 Sol as a 272K-context option on paid plans in the US and EU, with a 2.4x credit multiplier relative to Kiro Auto. The change matters because Sol is now selectable inside an engineering agent environment with its own planning, review, and execution workflow, rather than being discussed only as a standalone model capability.

Kiro describes that workflow as spec-driven development: prompts can be turned into requirements, architecture designs, and sequenced tasks before implementation. OpenAI says GPT-5.6 in Kiro can use requirements, codebase context, and team standards for longer-running work, while Kiro documents property-based testing as a way to check behavioral rules beyond ordinary unit-test examples. In practical terms, the integration is aimed at making model choice part of a structured software-delivery process, not merely a chat-based code-generation choice.

The Kiro comparison table makes the product tradeoff unusually explicit. Sol, Terra, and Luna each have 272K context, but their multipliers are 2.4x, 1.0x, and 0.1x respectively; Kiro recommends Sol for the hardest multi-step refactors and terminal work, Terra for routine agentic tasks, and Luna for high-throughput work. OpenAI and AWS report a roughly 82% cost reduction for GPT-5.6 Terra on Terminal-Bench 2.1 in their Kiro testing, but that result is about Terra rather than Sol and is not a public, model-to-model benchmark comparison for Sol.

Technical teams should therefore treat the Kiro availability as a configuration and evaluation opportunity, not proof that Sol is the right default. Sol is unavailable on Kiro's free tier, and Kiro says GPT-5.6 requests are served from the US regardless of a user's profile geography, which can matter for plan selection and data-processing requirements. Teams considering Sol should test representative repository tasks, review behavior, and credit consumption against Terra, Luna, and their existing preferred model before standardizing on it.