Which GPT-5.6 models can I pick in Kiro?

Kiro's model docs list three GPT-5.6 tiers — Sol, Terra and Luna — described as balancing agentic performance against cost. Kiro also offers an Auto option that routes a task to the model it judges best. The docs state that GPT-5.6 brings OpenAI models to Kiro for the first time.

Does GPT-5.6 actually make Kiro cheaper to run?

OpenAI and AWS report that on Terminal-Bench 2.1, GPT-5.6 Terra completed successful tasks in Kiro at roughly 82% lower cost. That figure comes from the two companies' own joint testing, not an independent evaluation, and it measures cost per completed task rather than a change in token pricing.

What other models does Kiro support besides GPT-5.6?

Kiro's documentation lists Claude Opus 5, 4.8, 4.7, 4.6 and 4.5, Claude Sonnet 5, 4.6, 4.5 and 4.0, Claude Haiku 4.5, DeepSeek 3.2, MiniMax M2.5 and M2.1, GLM-5, and Qwen3 Coder Next, alongside the new GPT-5.6 tiers. Kiro describes this as access to frontier and open-weight models from several providers.

Can I review the agent's work before changes are applied?

Yes. OpenAI says developers using GPT-5.6 in Kiro can review and refine the model's work at checkpoints before changes are implemented. Kiro can also check whether an implementation is correct using property-based testing, which generates many inputs against a stated property rather than relying on a fixed set of example cases.

OpenAI · 2026-08-24 · major

GPT-5.6 in Kiro — OpenAI's model family lands in AWS's coding agent

GPT-5.6 Sol, Terra and Luna are now available in Kiro, AWS's spec-driven coding agent. OpenAI and AWS say joint testing on Terminal-Bench 2.1 showed GPT-5.6 Terra finishing tasks in Kiro at roughly 82% lower cost.

OpenAI and Kiro lockup card for GPT-5.6 arriving in AWS Kiro

OpenAI's GPT-5.6 family is now selectable inside Kiro, AWS's spec-driven coding agent, for the first time.

Key specs

Cost per finished task	~82% lower on Terminal-Bench 2.1

Quick facts

Maker	OpenAI, with AWS
Models added	GPT-5.6 Sol, Terra and Luna
Host	Kiro, AWS's spec-driven development agent
What's new	First time OpenAI models are available in Kiro
Also in Kiro	Claude, DeepSeek 3.2, MiniMax M2.5, GLM-5, Qwen3 Coder Next
Availability	Live in Kiro now

What is it?

Kiro can now run OpenAI models. The GPT-5.6 family — Sol, Terra and Luna — is selectable inside AWS's agent, which Kiro's own docs call the first time OpenAI models are available there. Kiro is built around spec-driven development: it converts a high-level request into requirements, a technical design and a list of executable tasks before any code gets written.

How does it work?

The claimed gain comes from context, not a bigger model. Rather than a bare prompt, Kiro hands GPT-5.6 structured requirements, designs and task context drawn from the codebase and the team's standards, so the model starts out knowing what is being built and what the finished implementation has to do. Developers step in at checkpoints to review output before changes land, and Kiro can check correctness with property-based testing.

Why does it matter?

Cost per finished task is what engineering teams actually pay for, and that is the number OpenAI and AWS chose to report for GPT-5.6 in Kiro. For AWS shops the practical effect is choice: OpenAI models now sit next to Claude, GLM-5 and Qwen3 Coder Next in the same agent, so a team can match model to stage — cheap for scaffolding, stronger for a hard refactor — without leaving the tool.

Who is it for?

teams doing agentic coding on AWS

Frequently asked questions

Which GPT-5.6 models can I pick in Kiro?: Kiro's model docs list three GPT-5.6 tiers — Sol, Terra and Luna — described as balancing agentic performance against cost. Kiro also offers an Auto option that routes a task to the model it judges best. The docs state that GPT-5.6 brings OpenAI models to Kiro for the first time.
Does GPT-5.6 actually make Kiro cheaper to run?: OpenAI and AWS report that on Terminal-Bench 2.1, GPT-5.6 Terra completed successful tasks in Kiro at roughly 82% lower cost. That figure comes from the two companies' own joint testing, not an independent evaluation, and it measures cost per completed task rather than a change in token pricing.
What other models does Kiro support besides GPT-5.6?: Kiro's documentation lists Claude Opus 5, 4.8, 4.7, 4.6 and 4.5, Claude Sonnet 5, 4.6, 4.5 and 4.0, Claude Haiku 4.5, DeepSeek 3.2, MiniMax M2.5 and M2.1, GLM-5, and Qwen3 Coder Next, alongside the new GPT-5.6 tiers. Kiro describes this as access to frontier and open-weight models from several providers.
Can I review the agent's work before changes are applied?: Yes. OpenAI says developers using GPT-5.6 in Kiro can review and refine the model's work at checkpoints before changes are implemented. Kiro can also check whether an implementation is correct using property-based testing, which generates many inputs against a stated property rather than relying on a fixed set of example cases.

Try it

Pick GPT-5.6 Sol, Terra or Luna from the model picker in Kiro — https://kiro.dev/docs/models/