Kiro vs Claude Code: credits, plans, Claude in Kiro, and spec-driven work

Kiro is AWS's agentic IDE, built on Code OSS, with a CLI alongside; it runs Claude, GPT and open-weight models and is built around specs that it writes before the code. Claude Code is Anthropic's agent and runs Claude only. Both start at $20 a month, but they meter differently: Kiro counts credits per request, scaled by the model, and Claude Code a five-hour window with a weekly limit.

Kiro pricing vs Claude Code pricing

Kiro has five plans, each per user: Free with 50 credits a month, Pro at $20 with 1,000, Pro+ at $40 with 2,000, Pro Max at $100 with 5,000 and Power at $200 with 10,000. On every paid plan extra credits cost $0.04 each, and Kiro credits $20 when you first upgrade from Free with a social login or Builder ID. Credits are charged in fractions per request and billed on the 1st of the month.

Claude Code is included in every paid Claude plan: Pro at $20, or $17 billed annually; Max at $100 or $200 for five or twenty times Pro's usage. The details: Claude Code pricing.

What Claude costs in Kiro

A Kiro request costs credits times the model's multiplier. Auto, which picks the model for you, is 1×. Claude Opus 5.5 is 2×, Opus 5 and 4.8 are 2.2×, Sonnet 5.5 and 5 are 1.3×, and Haiku 4.5 is 0.4×. Kiro's pricing page gives the free plan Claude Sonnet 4.5 and open-weight models, and premium models from Pro up.

That makes Kiro and Claude Code hard to compare in dollars: Kiro does not publish how many credits a typical request takes, and Anthropic does not publish how many tokens a plan's window holds. What I can measure is the Claude Code side: my median active day came to $153 at Anthropic's list prices, over 56 active days from 18 July to 2 October 2026. Caprock does not read Kiro, so there is no Kiro figure here, measured or estimated.

Spec-driven work vs plan mode

Kiro's specs turn a prompt into three files before any code: requirements.md, design.md and tasks.md, which you approve step by step; a bug fix uses bugfix.md in place of requirements. It runs independent tasks in parallel. Steering files, hooks on file changes and tool use, and MCP servers shape what the agent does.

Claude Code's nearest equivalent is plan mode: Claude researches and writes a plan you approve before it edits. CLAUDE.md, hooks, subagents and MCP play the parts of steering, hooks and tools; there is no fixed three-file format.

Kiro CLI vs Claude Code

Kiro CLI replaced the Amazon Q Developer CLI in November 2025, and the old q commands still work. It has a headless mode, custom agents and subagents, MCP, steering and hooks:

The Kiro IDE runs on macOS, Windows and Linux, and takes your VS Code settings and Open VSX extensions. You sign in with GitHub, Google, an AWS Builder ID or IAM Identity Center; no AWS account is needed. Claude Code runs in a terminal, in VS Code and JetBrains, in a desktop app and at claude.ai/code.

Which one to pick

• Kiro — you want the agent to write a spec you approve before it codes, in an IDE, and to choose between Claude and cheaper open-weight models per request. The free plan lets you try it.
• Claude Code — you want Claude, in the terminal or your own editor, for hours a day on a flat fee inside its limits.
• Both — Kiro's credits and a Claude plan are separate bills.

Cursor vs Claude Code and GitHub Copilot vs Claude Code compare two other editors.

See your Claude side

Caprock reads Claude Code's transcripts from disk, whether or not it started the session, and prices every call — per session, per project, per model — a list-price estimate as Claude Code's own /usage is, next to the plan's dollars and the share of the weekly limit the work used. It does not read Kiro.

Free, open source; nothing leaves your machine. Install Caprock and it reads back every Claude Code session still on disk.