GPT-5.6 Family Lands in Amazon’s Kiro Coding Agent

TLDR

OpenAI on Monday, August 24, 2026 (Product) put the GPT‑5.6 family — Sol, Terra, and Luna — into Kiro, Amazon’s spec-driven software-development agent. Developers can plan, build, review, and test against requirements, designs, and repo context; Kiro adds checkpoint review and property-based testing. OpenAI × AWS testing: on Terminal-Bench 2.1, GPT‑5.6 Terra completed successful tasks in Kiro at roughly an 82% cost reduction. Quotes: Swami Sivasubramanian (AWS VP, Agentic AI) and Colleen Kapase (OpenAI VP, Strategic Global Partnerships).

What shipped

Item	openai.com/index Product Aug 24
Models	GPT‑5.6 Sol, Terra, Luna in Kiro
Partner	AWS — environment + model optimization
Claimed bench	Terminal-Bench 2.1; Terra ~82% cost reduction in Kiro
Workflow	Specs → tasks → checkpointed implementation → property tests

This is IDE/agent distribution, not a new base model. Analogous to Grok 4.6 in GitHub Copilot (Aug 14) and Daybreak on Bedrock (Aug 11). The 82% figure is joint vendor testing of Terra inside Kiro’s harness, not an independent AA/Arena rank of a new checkpoint.

Product-line de-dupe: not Jul 9 GPT‑5.6 family GA, not Jul 30 Luna/Terra price cuts, not Aug 6 Sol ChatGPT retune, not Aug 13 Ultrafast Sol / Cerebras.

Why this story matters

Kiro is AWS’s answer to Cursor-class agents. Putting Sol/Terra/Luna on that surface is how OpenAI keeps enterprise coding from defaulting to Claude or Grok inside Amazon’s own tool. Watch: whether Kiro exposes Ultrafast Sol, whether the 82% holds on public Terminal-Bench leaderboards, and Codex vs Kiro cannibalization.

Sources

• OpenAI: Advancing price-performance for developers with GPT‑5.6 in Kiro (August 24, 2026)
• Times of AI: spacexai-grok-4-6-github-copilot (August 14); openai-ultrafast-gpt-5-6-sol-cerebras (August 13)