OpenAI’s GPT-5.6 model family is now available in Kiro, the AWS agentic coding service built for spec-driven development and testing.
The update brings three models from OpenAI’s latest flagship series – Sol, Terra, and Luna – into Kiro. AWS and OpenAI describe the pairing as a way to help developers produce code with fewer iterations and better cost-per-token outcomes.
Swami Sivasubramanian, VP of Agentic AI at AWS, said: “We are always looking to make the latest foundation models available to developers and expand their options to accelerate AI-native development using Kiro. We are excited to add the GPT‑5.6 family of models to power complex and long-running development tasks in Kiro.”
Property-based testing checks correctness before code ships
Kiro converts product ideas and requirements into implementation plans as a first step, rather than starting from an open-ended prompt. From there, the agent works through multi-step coding tasks while pulling context from across a team’s codebase and its established engineering standards.
Developers can review and refine the model’s output at set checkpoints before any change is implemented. Once implementation is complete, Kiro checks its correctness using property-based testing, a method that verifies code behaviour against defined properties rather than a fixed set of example inputs.
That combination puts an automated correctness check and a manual review gate on either side of the model’s output.
OpenAI and AWS say they have optimised the Kiro environment for the OpenAI models it now runs. On Terminal-Bench 2.1 – a benchmark that scores whether an AI agent completes assigned terminal-based tasks correctly – testing found that GPT-5.6 Terra completed successful tasks in Kiro at roughly an 82 percent cost reduction, the companies said.
Kiro’s spec-driven approach grounds the model in requirements, technical designs, and task context from the outset. OpenAI and AWS attribute the model’s ability to reach a working solution with fewer missteps to that grounding.
Colleen Kapase, VP of Strategic Global Partnerships and Ecosystems at OpenAI, commented: “By bringing the GPT‑5.6 family to Kiro, developers gain more room to match intelligence, speed, and cost to each stage of the software development lifecycle.
“Together with AWS, we’re helping teams get more from every dollar they invest in AI while moving faster when time matters.”
Review checkpoints keep a manual gate in the process
Kiro’s checkpoint reviews give developers a manual gate before code changes land, sitting alongside the automated property-based tests as a second control point in the same workflow. Developers can accept, adjust, or reject the model’s proposed changes at those checkpoints before implementation proceeds.
The two documented controls – checkpoint review and property-based testing – represent the mechanisms Kiro applies before treating generated code as finished. Cost and speed claims tied to Terminal-Bench 2.1 describe benchmark performance under test conditions. However, what they don’t describe are defect rates, rework volumes, or production incident data from teams running GPT-5.6 inside Kiro.
OpenAI and AWS say they will keep working together to improve how OpenAI’s models perform inside Kiro and to help developers get more value from AI across the software development lifecycle. The GPT-5.6 family is available in Kiro now.
See also: AWS DevOps Agent traces pipeline failures to GitHub commits
Want to learn more about cybersecurity from industry leaders? Check out Cyber Security & Cloud Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the AI & Big Data Expo. Click here for more information.
Developer is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.