Disclosure: some links on this page may be affiliate links; we may earn a commission at no extra cost to you. Read the disclosure.

By the toolist.ai editors. Researched from Kiro's own pages and hands-on tests published by others, not hands-on tested by toolist.ai. Updated October 20, 2026.

Kiro review: pricing, features and what testers found

Kiro is Amazon Web Services' AI coding agent: an IDE built on Code OSS (the open-source base of VS Code), a terminal CLI, a browser agent (Kiro Web) and the open-source Kiro Crew workspace, built around specs that turn a prompt into requirements, a design and a task list before agents write the code. Free gives 50 credits a month; Pro costs $20 a user a month for 1,000 credits.

Our take

Kiro suits developers and teams who want an AI agent to plan before it codes: its specs turn a prompt into requirements, a design and a task list, and the same agent works in a VS Code-style IDE, a terminal CLI, the browser (Kiro Web, on paid plans) and the open-source Kiro Crew. Free gives 50 credits a month on Auto, Claude Sonnet 4.5 and open-weight models; Pro costs $20 a user a month for 1,000 credits and adds models such as Claude Opus 5.5 and GPT-5.6, and extra credits cost $0.04 each. Testing the IDE in 2026, HostAdvice's Walter Akolo got a plan of 43 sub-tasks traced back to 12 requirements for a property management platform and scored Kiro 8.2 out of 10, but the planning alone used 4.28 credits and one step timed out after 7 minutes 24 seconds. Tests of the 2025 preview found the same structure with rough edges: The New Stack's David Eastman judged the spec approach strong on his Rails app but thought the workflow may need simplifying, and InfoWorld's Serdar Yegulalp had to step in constantly, hit API timeouts and syntax errors, and got 230 lines for a tool whose basic version would need no more than two dozen. The catches are the credit meter, which charges more for stronger models (Claude Opus 5.5 uses twice Auto's credits, GPT-5.6 Sol 4.4 times) and does not roll monthly credits over, and data use: on Free and on paid plans you sign into with GitHub, Google or AWS Builder ID, Kiro may use your content for service improvement, including model training, unless you opt out. This verdict rests on those three published tests and Kiro's own pages.

Based on Kiro's published plans and the hands-on tests linked below, not on a test of our own.

Good for

  • Developers and teams who want the agent to write requirements, a design and a task list they can review and edit before any code is written, especially for larger features
  • Teams on AWS that want sign-in through IAM Identity Center or their own identity provider, billing on their AWS account and content kept out of service improvement
  • Handing long tasks to cloud agents that open pull requests across GitHub and GitLab repositories from the browser or a phone, on paid plans
  • Trying a spec-driven coding agent at no cost on the free tier, or for a year as a verified student at a listed university (1,000 credits a month, no credit card)

Watch out for

  • Everything is metered in credits, which Kiro defines as a unit of work: a simple prompt can cost less than 1 credit, a spec task usually more, and the model sets the rate (Claude Opus 5.5 uses twice Auto's credits, GPT-5.6 Sol 4.4 times, doubled again for GPT-5.6 requests over 272K tokens), so a plan's credits do not translate into a fixed number of tasks. In HostAdvice's 2026 test, planning one platform used 4.28 credits before any application code was written. Unused monthly credits do not roll over.
  • Past the monthly credits, individual paid subscribers buy add-on credits at $0.04 each, in packs from $5 (125 credits) to $100 and up to five packs at a time, which expire 12 months after purchase. That is twice the 2 cents a credit every paid plan works out to, and Kiro itself suggests moving up a plan once add-on spend nears the price difference. Free cannot buy add-ons and pauses until the next month; team overages ($0.04 a credit, billed at month end) stay off until an admin enables them.
  • Billing runs on calendar months from the 1st (UTC). Your first upgrade is prorated, and the $20 sign-up credit applies only when you sign in with GitHub, Google or AWS Builder ID, not through IAM Identity Center; from your second paid month, mid-month upgrades charge the full price difference for the month, while downgrades and cancellations take effect the next month after you pay the current one in full. Kiro lists no yearly billing, and its FAQ names only credit cards as a payment method.
  • On Free and on paid plans you sign into with GitHub, Google or AWS Builder ID, Kiro may use your questions, inputs, its responses and generated code for service improvement, which its docs say can include model training, unless you untick content collection under Settings, Application, Telemetry and Content; free-tier inputs may also be stored for up to 60 days for abuse detection. Kiro's enterprise page says content on Pro to Power subscriptions is not used to train foundation models, without naming a sign-in method, while its FAQ and data-protection page keep only users who sign in through IAM Identity Center or an external identity provider out of service improvement.
  • Kiro's output indemnity on paid plans does not cover its Excluded Models: Kimi K3 and every model marked experimental, preview, pre-release or beta; Kiro's model table marks Claude Opus 5.5, all three GPT-5.6 models and the open-weight DeepSeek 3.2, MiniMax M2.1 and Qwen3 Coder Next experimental. Model availability also varies by country, and free and individual users' requests go to Kiro's US endpoint, with their content stored in US East (N. Virginia).
  • In the IDE, Autopilot is the default: the agent creates files, changes code across the project, runs commands and makes architectural decisions without asking at each step (Supervised mode stops for approval after each turn with edits), and the local command sandbox that limits what agent-run commands can read, write and reach is off by default.
  • Your subscription may be used only in Kiro's own apps, ACP-compatible editors and software-development automation such as CI reviews; routing it through third-party harnesses such as OpenClaw is not permitted. Each developer needs their own subscription.
  • Two of the three hands-on tests cited here cover the 2025 preview, before general availability in November 2025, the CLI, Kiro Web, Crew and today's models, when InfoWorld's tester hit API timeouts and over-long code; HostAdvice's 2026 tester still hit a timeout after more than 7 minutes. Older reviews also quote preview-era plans ($19 and $39 a month) and separate vibe and spec request limits, which Kiro replaced with one credit pool in September 2025, or the 500 bonus credits new users got until May 8, 2026, when Kiro replaced them with the $20 credit on a first paid upgrade.

What Kiro does

Price

Free plan: Kiro Free gives 50 credits a month with no credit card, on Auto, Claude Sonnet 4.5 and open-weight models such as Qwen3 Coder Next, DeepSeek 3.2 and MiniMax M2.1, subject to rate limits. It cannot buy add-on credits, so work pauses until the next month's reset, and Kiro Web, cloud sessions, the iOS app and API keys for headless CLI runs need a paid plan. Kiro may use your content for service improvement unless you opt out, and may keep free-tier inputs for up to 60 days for abuse detection.

PlanBilled monthlyBilled yearly, per monthIncludes
Free free not listed 50 credits a month, no credit card; Auto, Claude Sonnet 4.5 and open-weight models (DeepSeek 3.2, MiniMax M2.5 and M2.1, GLM-5 and Qwen3 Coder Next), subject to rate limits; no add-on credits, Kiro Web, cloud sessions, iOS app or headless API keys; content may be used for service improvement unless you opt out
Pro $20 not listed Price per user: 1,000 credits a month (unused credits do not roll over); premium models such as Claude Opus 5.5 and Sonnet 5.5, GPT-5.6 Sol, Terra and Luna and Kimi K3, where available in your country; Kiro Web, cloud sessions, the iOS preview and API keys for headless CLI runs; add-on credits at $0.04 each; output indemnity outside Kiro's Excluded Models; a $20 credit, prorated by the days left in the month, the first time you upgrade with GitHub, Google or AWS Builder ID sign-in
Pro+ $40 not listed Price per user: 2,000 credits a month; otherwise as Pro
Pro Max $100 not listed Price per user: 5,000 credits a month; otherwise as Pro (added on June 11, 2026, between Pro+ and Power)
Power $200 not listed Price per user: 10,000 credits a month; otherwise as Pro
Teams and enterprise (through AWS) not listed not listed The same Pro, Pro+, Pro Max and Power credits and per-user prices for teams of 2 to 500, subscribed by an admin with AWS IAM Identity Center or an external identity provider (Okta, Microsoft Entra ID) and billed monthly on the AWS bill; overages at $0.04 a credit, billed at month end, off until an admin enables them and capped through AWS Service Quotas; no free tier and no $20 sign-up credit; content not used for service improvement; about 20% higher in AWS GovCloud (US); more than 500 users or custom terms through sales

Prices last checked: 2026-10-30 on Kiro's pricing page. Full breakdown: Kiro pricing.

Price history

No price change since we first recorded these prices on October 20, 2026. We re-check them every week and list each change here.

What hands-on testers found

We have not tested Kiro ourselves. These are the published tests we found that used it hands-on, newest first; follow the links for the full reviews.

HostAdvice · Walter Akolo · updated September 24, 2026

What they tested: A property management platform for landlords and tenants, built in Spec mode in the Kiro IDE on Linux (Pop!_OS): steering documents, requirements, design and task list, then the first scaffolding task, with Autopilot on for planning and manual approval for the task. First published June 30, 2026 (the page's metadata), with screenshot files dated June 2026; his plan is not stated, though he measures the session against the free tier's 50 credits.

  • From one prompt Kiro wrote 12 requirements, refined them with parallel sub-agents, then produced a design and a plan of 11 task groups and 43 sub-tasks, each with commands, file paths and references back to the requirements, detailed enough in his view for a human developer to work from.
  • The planning alone used 4.28 credits, and the chat showed an estimated credit cost and time after each agent action; he judged the free tier's 50 credits a genuine way to evaluate Kiro but short for full spec sessions on anything complex.
  • One request timed out after 7 minutes 24 seconds between the requirements and design steps; Kiro picked up again without losing the finished requirements, and his score of 8.2 is held back by the free tier and that failure.

Their score: 8.2 out of 10

The New Stack · David Eastman · September 20, 2025

What they tested: Spec mode in the Kiro IDE, a few weeks after he got off the waitlist (the article ran on September 20, 2025) and before general availability, the CLI and today's models, with the model set to Claude Sonnet 4: adding create, read, update and delete for a 'Thought for the Day' linked to an existing model in his Rails game-design app, through requirements, design and the first task. The page notes that Amazon Web Services is a sponsor of The New Stack; the article is not marked as sponsored.

  • Kiro found the one related model in his project and wrote six requirement sets that caught the key link to it and that text was required but an image was not, though it cast the user as an administrator; he felt it covered what mattered with little help from him.
  • The design ran to about 170 lines and picked up his Bootstrap styling, and every task traced back to a requirement; the first task's migration stored the link as a plain integer rather than a Rails reference, the kind of detail he expects to argue over with an AI.
  • He judged the spec approach strong and aimed at teams who review the documents outside the tool, but thought the workflow may need simplifying, since editing one requirement updates the whole tree, and feared loops between hooks, which he did not test, could make it brittle.

InfoWorld · Serdar Yegulalp · updated July 30, 2025

What they tested: The Kiro preview on Windows in July 2025, during its open preview and before pricing, general availability, the CLI and today's models, with Claude Sonnet 3.7 or 4.0: a vibe-mode Python tool that finds broken virtual environments, and a spec-mode command-line static site generator, watching each step instead of using Autopilot. First published July 22, 2025.

  • The spec flow produced WHEN/THEN user-story requirements, a design document and an interactive task list with live feedback at each step, but he had to step in constantly; an instruction to run Python as py on Windows held only once, even after he added it to the steering files.
  • Generated code and tests often had syntax errors a linting step would have caught, fixing a hanging test took several rewrites and a restart of the step, and the finished generator still lacked the templates its own tests called for.
  • The virtual-environment checker ran to 230 lines with useful options he had not asked for, such as JSON export, where a basic version would need no more than two dozen lines; frequent Claude API timeouts made the preview bumpy, and he suggested looking again once AWS had added capacity.

What users say

Test result

toolist.ai has not run its own test of Kiro; the findings above come from the published tests linked. See how we research and test.

Best coding tools · Kiro pricing

Do you make Kiro? Claim this listing to correct it; it's free. To see whether ChatGPT, Claude, Gemini and Perplexity recommend it, and how many readers we send you: toolist.ai for AI companies.

Sources