Kiro, the AWS-powered coding platform, just rolled out OpenAI’s GPT-5.6 model family across its entire product stack. The integration spans Kiro’s IDE, CLI, and web interface, and it marks the first time the platform has adopted OpenAI models after building its foundation almost exclusively on Anthropic’s Claude variants.

What Kiro actually shipped

The GPT-5.6 rollout introduces three model variants: Sol, Terra, and Luna. Each is designed for a different spot on the performance-cost spectrum.

Sol is the heavy hitter, built for complex agentic coding tasks. It carries a 2.4x credit multiplier, meaning it burns through user credits faster but delivers top-tier results. Sol scored an 80 on the Coding Agent Index and 88.8% on Terminal-Bench 2.1.

Terra sits in the middle at a 1.2x multiplier, offering balanced performance for everyday development work. Luna, at 0.6x, is the speed-and-affordability play for lighter tasks where raw power isn’t the priority.

All three variants share a 272K context window. The models are being served through Amazon Bedrock infrastructure with an “Auto” routing mode that directs queries to the optimal variant based on task complexity.

The rollout is experimental for now, available to Pro, Pro+, Pro Max, and Power tier subscribers. Geographic availability is limited to the us-east-1 and eu-central-1 regions.