# Kiro vs Claude Code in 2026: specs or a general-purpose agent?

Kiro writes a spec before it writes code and bills through AWS. Claude Code is Anthropic's own agent. How they compare, and where each fits.

Published 2026-10-07 by Richard Kaminsky and Mitchell Lipyansky, the co-founders of Poly. Canonical: https://usepoly.co/kiro-vs-claude-code
Poly is a multiplayer AI coding workspace: a shared room where your team works with one AI agent, together. Free to start: https://usepoly.co/

**Kiro is AWS's agentic IDE and command-line tool, built around specs: before writing code it drafts requirements, a design and a task list you can review. Claude Code is Anthropic's agent, which works directly in your terminal, editor or browser and runs Claude models only. Both start at $20 a month and both run Claude, but Kiro meters usage in credits, where Opus 5.5 costs twice the default and Sonnet 5.5 1.3 times, while Claude Code draws on a Claude plan's usage limits. Claude Code leads the independent agent benchmarks; Kiro isn't listed on them.**

## Key takeaways

- **Pricing:** Kiro is free with 50 credits a month, then Pro $20 (1,000 credits), Pro+ $40, Pro Max $100 and Power $200. Claude Code starts with Claude Pro at $20.
- **Models:** Kiro runs Claude, OpenAI's GPT-5.6 and open-weight models, with per-model credit multipliers. Claude Code runs Claude only.
- **Workflow:** Kiro's specs, steering files and hooks against Claude Code's plan mode, CLAUDE.md, skills and hooks.
- **Teams:** Kiro bills through AWS with IAM Identity Center, Okta or Entra ID, and is FedRAMP High in GovCloud. Claude Code comes with Claude Team and Enterprise.
- **Together:** Kiro Crew, Kiro's open-source always-on workspace, can run Claude Code as its agent in preview.

*Kiro and Claude Code, from Kiro's and Anthropic's pricing pages and docs, October 7, 2026.*

|  | Kiro | Claude Code |
| --- | --- | --- |
| Made by | AWS | Anthropic |
| Approach | Spec-driven: requirements, design, tasks | General-purpose agent with plan mode |
| Models | Claude, GPT-5.6 and open-weight models | Claude |
| Free plan | 50 credits a month | None |
| Paid from | Pro, $20 (1,000 credits) | Claude Pro, $20 ($17 annually) |
| Metering | Credits with per-model multipliers | Five-hour and weekly limits |
| Where it runs | Kiro IDE, CLI, web, iPhone, other IDEs via ACP | Terminal, VS Code, JetBrains, desktop, web, phone |
| Enterprise | Through AWS; FedRAMP High in GovCloud | Claude Team and Enterprise |
| Independent agent benchmarks | Not listed | Leads Terminal-Bench 4.0 and Artificial Analysis |

## What is Kiro?

AWS launched Kiro in preview in July 2025 as an IDE built on Code OSS, the open-source base of VS Code, and made it generally available in November 2025 alongside a CLI that replaces the Amazon Q Developer CLI ([Kiro](https://kiro.dev/blog/general-availability/)). It now also runs in a browser (Kiro Web) and on iPhone, both in preview, and all of them share one agent.

Its defining feature is **specs**: for a feature, Kiro writes `requirements.md`, `design.md` and `tasks.md` (or a bug-fix spec), which you review before it builds ([Kiro](https://kiro.dev/docs/specs/)). Steering files hold project instructions, hooks run commands or prompts on events such as saving a file, and it supports MCP, skills and subagents. In Autopilot, the default, it works through tasks on its own; in Supervised mode it stops for your approval after each turn that edits files.

## How much do Kiro and Claude Code cost?

**Kiro** ([pricing](https://kiro.dev/pricing/)):

- **Free:** 50 credits a month.
- **Pro, $20:** 1,000 credits. **Pro+, $40:** 2,000. **Pro Max, $100:** 5,000. **Power, $200:** 10,000.
- **Overage:** $0.04 a credit.

A credit is a unit of work: simple prompts cost under one, spec tasks more. Each model has a multiplier against Kiro's Auto mode: Sonnet 5.5 costs 1.3 times, Opus 5.5 2 times, Fable 5.1 6 times (enterprise preview only), and open-weight models as little as 0.05 times ([Kiro](https://kiro.dev/docs/models/)). Opus 5.5 and Sonnet 5.5 are marked experimental and are on paid plans only.

**Claude Code** comes with Claude plans: Pro at $20 a month ($17 billed annually), Max at $100 or $200, Team seats from $20, and Enterprise at $20 a seat plus usage. Usage is a five-hour and weekly allowance shared with Claude chat ([Claude Code usage limits](/claude-code-usage-limits)).

## How do their workflows compare?

- **Planning:** Kiro's specs are written files you can edit and keep in the repository. Claude Code's plan mode produces a plan in the conversation; you can ask it to write one to a file.
- **Project instructions:** Kiro's steering files and AGENTS.md; Claude Code's CLAUDE.md and skills.
- **Automation:** both have hooks that run on events, MCP servers and subagents.
- **Approval:** Kiro's Supervised mode asks after each editing turn; Claude Code asks before edits and commands by default, with modes that ask less.
- **Where it runs:** Kiro in its own IDE, CLI, browser and phone, and as an agent inside JetBrains, Zed and others. Claude Code in the terminal, VS Code, JetBrains, its desktop app, browser and phone.

## Which is better on benchmarks?

Kiro doesn't appear on Terminal-Bench 4.0, Artificial Analysis's coding-agent index or SWE-rebench. Claude Code leads the first two: 64.8% on Terminal-Bench with Opus 5.5 and 68 on Artificial Analysis with Sonnet 5.5 ([the rankings](/best-ai-coding-agents)). Kiro's documentation quotes scores for the models it offers, not for Kiro itself, so they don't compare.

## How do they work for teams?

**Kiro** sells the same per-user plans to teams of up to 500 and, for larger organizations, through the AWS console: sign-in through IAM Identity Center, Okta or Entra ID, usage on the AWS bill, overage off unless an admin enables it. It is FedRAMP High and DoD IL-4/5 authorized in AWS GovCloud. Kiro doesn't use content from users who sign in through an organization's identity provider; free and personal sign-ins may be used to improve the service unless they opt out.

**Claude Code** comes with Claude Team ($20 to $125 a seat) and Enterprise ($20 a seat plus usage), with SSO, and SCIM and audit logs on Enterprise ([Claude Team vs Enterprise](/claude-team-vs-enterprise)).

Neither lets several people share one live session; teams share specs, steering files and pull requests through git.

## Can you use Kiro and Claude Code together?

Yes, in two ways. Kiro Crew, an open-source always-on workspace Kiro launched in August 2026, can use Claude Code as its agent backend in a developer-mode preview ([Kiro](https://kiro.dev/docs/crew/features/agent-backends/)). And both follow the open Agent Skills format, so skills can move between them.

## Kiro vs Cursor

Both are VS Code-family editors. Kiro's edge is specs, AWS billing and compliance, cheap open-weight models and team seats at the individual price. Cursor's is a mature editor with cloud agents, Bugbot reviews and a team marketplace, at $40 a seat for teams ([Cursor for teams](/cursor-for-teams)).

## Which should you choose?

- **You want the strongest agent and a flat Claude plan:** Claude Code.
- **You want to review a written spec before code gets written, or your company buys through AWS:** Kiro.
- **You need FedRAMP High or GovCloud:** Kiro.
- **A team that wants to steer one agent together:** Poly (usepoly.co) runs Claude Code and Codex in a shared browser room where everyone sees each step and any member can approve a change before it happens. Free to start. [What is Poly?](/what-is-poly)

## Common questions

**Is Kiro better than Claude Code?**

Claude Code leads the independent agent benchmarks, where Kiro isn't listed. Kiro's strengths are its spec-driven workflow, AWS billing and compliance, and cheaper open-weight models.

**Does Kiro use Claude?**

Yes. Kiro offers Claude models including Sonnet 5.5 and Opus 5.5 on paid plans, at 1.3 and 2 times the credit cost of its Auto mode, alongside GPT-5.6 and open-weight models.

**How much does Kiro cost?**

Kiro is free with 50 credits a month. Paid plans are Pro at $20 for 1,000 credits, Pro+ at $40, Pro Max at $100 and Power at $200, with extra credits at $0.04 each.

**What is spec-driven development in Kiro?**

Before building a feature, Kiro writes a requirements document, a design and a task list, which you review and edit. It then works through the tasks.

**Can Kiro run Claude Code?**

Kiro Crew, Kiro's open-source always-on workspace, can use Claude Code as its agent backend in a developer-mode preview. The Kiro IDE runs its own agent.

More guides: https://usepoly.co/guides · Security: https://usepoly.co/security · Pricing: https://usepoly.co/pricing