# Kiro Pricing Explained: How Far $20 a Month Goes

> Kiro gives a free tier of 50 credits and Pro at $20 a month for 1,000. What a credit actually buys, the Auto versus pinned-model savings, and how overage works.

Source: https://capitalandcompute.net/blog/kiro-pricing-explained-free-tier-credits/
Published: 2026-07-01
Updated: 2026-10-12

__Kiro's monthly credit allotment, by plan__
| Item           | Value  |
| -------------- | ------ |
| Free ($0)      | 50     |
| Pro ($20)      | 1,000  |
| Pro+ ($40)     | 2,000  |
| Pro Max ($100) | 5,000  |
| Power ($200)   | 10,000 |

Kiro's credit ladder: Free gives 50 credits perpetually, and every paid tier scales in a straight line from there, Pro at 1,000 up to Power at 10,000\. The price per credit is identical at every paid tier, about two cents.Source: [Kiro (Cognition/AWS), Pricing, 2026](https://kiro.dev/pricing/)

Fifty credits, forever, for free. Then $20 buys 1,000\. Every tier above that is the same math: double the price, double the credits. Kiro’s [pricing page](https://kiro.dev/pricing/) is, refreshingly, one of the more literal pricing pages in AI coding tools right now. No hidden request caps, no separate limits for different task types. One number, one pool.

The part that page does not spell out is what a credit is actually worth in real work, and there is one lever inside Kiro that changes that answer by close to a quarter. That is the part worth understanding before you pick a tier.

Key takeaways

* Free is 50 credits a month, permanently, with access to open-weight models and Claude Sonnet 4.5, rate-limited. Pro is $20/mo for 1,000 credits; Pro+ $40/mo for 2,000; Pro Max $100/mo for 5,000; Power $200/mo for 10,000.
* A credit is not a fixed unit. Kiro consumes credits fractionally, in 0.01 increments, based on how complex the request is. There is no published table of "X credits per task."
* Kiro's default agent, Auto, costs 1x credits. Pinning Claude Sonnet by hand instead costs 1.3x for the same work, about 23% cheaper per task on Auto, which means a fixed credit pool stretches about 30% further.
* Overage is opt-in, not automatic. Paid tiers can enable extra credits at $0.04 each, billed at the end of the cycle, but only if you turn that on. Leave it off and Kiro just stops instead of quietly running up a bill.
* First paid upgrade gets a $20 credit toward the subscription via social login or an AWS Builder ID, which covers a full month of Pro on its own.

## How much does Kiro cost?

Kiro Free is $0/mo for 50 credits. Every paid tier costs $0.02 per credit, so the price scales in a straight line with the credit pool:

| Plan    | Price / month | Credits | Price per credit |
| ------- | ------------- | ------- | ---------------- |
| Free    | $0            | 50      | n/a (free)       |
| Pro     | $20           | 1,000   | $0.02            |
| Pro+    | $40           | 2,000   | $0.02            |
| Pro Max | $100          | 5,000   | $0.02            |
| Power   | $200          | 10,000  | $0.02            |

Overage on any paid tier, if enabled, is $0.04 per additional credit, twice the in-plan rate.

## What a credit actually buys

Here is the honest answer: Kiro does not say. [Kiro’s own billing documentation](https://kiro.dev/docs/billing/) states that “credits are consumed fractionally based on each request. Simple edits and shorter prompts will use fewer credits than complex, lengthy tasks.” That is the entire specification. No per-action table, no “one credit equals one prompt,” no fixed conversion rate the way an older tool might have sold “500 prompts a month.”

That is a deliberate design, not an oversight. Kiro’s [September 2025 pricing announcement](https://kiro.dev/blog/new-pricing-plans-and-auto/) explains the reasoning: the company used to run separate limits for “vibe” prompts and “spec” tasks, and merged them into one credit pool with fractional, 0.01-increment consumption specifically so a quick edit and a long autonomous run would not compete for the same fixed slot count. The tradeoff is that nobody outside Cognition can hand you a clean “$20 buys N tasks” number, because the meter moves with the work, not with the request.

What this means for picking a tier

Do not size a Kiro plan by imagining a task count. Size it by trying the free 50 credits on your actual workload for a week, watching how fast the pool drains, and scaling from there. The 50 to 1,000 to 2,000 to 5,000 to 10,000 ladder is linear, so whatever rate you observe on Free scales cleanly to any paid tier.

## The one setting that changes your credit math

This is the number most Kiro pricing coverage skips, and it is the one genuinely actionable lever in the whole system.

Kiro’s default agent is called Auto. It mixes frontier and specialized models with routing and caching, and [Cognition’s own announcement](https://kiro.dev/blog/new-pricing-plans-and-auto/) states a task through Auto costs 1x credits. Pin Claude Sonnet by name instead, so every request goes to that one model whether or not it is the right size for the job, and the same task costs 1.3x credits: Auto is about 23% cheaper per task ((1.3 - 1.0) / 1.3). Flip that ratio around and it means the same fixed credit pool covers about 30% more tasks (1.3 / 1.0) when left on Auto instead of pinned to Sonnet.

1.0x

credits via Auto

Kiro's default, model-routed

1.3x

credits pinning Sonnet

same task, one model forced

\~30%

more tasks on Auto

same credit pool, 1.3 / 1.0

Run that across a 1,000-credit Pro plan and the difference is not trivial: at the Auto rate, 1,000 credits cover roughly 30% more work than the same pool spent entirely on a hand-pinned Sonnet. Most people pin a model out of habit, because that is how the older generation of coding tools worked: pick your model, live with it. Kiro’s own pricing rewards not doing that.

The instinct to pin a specific model usually comes from wanting predictable output quality. That is a fair trade to make deliberately. It stops being a fair trade when it happens by default, and the user never notices Auto was the cheaper option the whole time.

## What happens when you run out

Most subscription AI tools handle running out of quota one of two ways: hard stop, or automatic metered billing the moment you cross the line. Kiro does neither by default.

[Kiro’s billing documentation](https://kiro.dev/docs/billing/) describes overage as opt-in: paid tiers “allow you to opt in to overages,” priced at $0.04 per additional credit, and if you have opted in and used them, “you are charged the total overage fees for the month at the end of the current billing cycle.” Leave that setting off, which is the default, and running out simply stops the agent rather than generating a surprise line item.

That is a meaningfully more conservative design than [Devin Desktop’s Pro plan](/blog/windsurf-devin-desktop-rebrand-cost/), which bundles a metered autonomous-agent product into its flat monthly price with overage billed automatically at an unpublished rate the moment the included quota is spent. Kiro’s version puts the decision in the user’s hands before any extra dollar moves.

If you do turn overage on

Opting in removes the safety net. At $0.04 a credit, a task that would have cost 5 credits inside your plan costs the same 20 cents if you overrun, which is cheap in isolation and easy to lose track of at volume. Turn it on only once you have a sense, from a month of normal use, of how many credits your actual workload burns.

## Is Kiro worth $20 a month?

For an individual developer testing whether an agentic IDE fits their workflow, yes, and the free tier is the reason. Fifty credits a month, with no expiration and no card required, is enough to run Kiro against a real side project and see how the credit meter behaves on your own code, not a vendor’s demo. Compare that against [Cursor’s free Hobby tier](/blog/cursor-pricing-limits-explained/) or [GitHub Copilot’s entry plan](/blog/github-copilot-pricing-ai-credits/), and Kiro’s free tier is closer to a genuine trial than a crippled teaser.

Once the free tier is not enough, Pro’s $20 for 1,000 credits is the same price point as most of the field. The number that should decide the tier is not the sticker; it is whether the plan on Auto covers a normal month at the rate observed during the free trial. Pro+ and above exist for exactly the people who did that math and found 1,000 credits too thin.

The [2026 AI coding agent landscape](/blog/the-2026-ai-coding-agent-landscape/) covers where Kiro sits against the rest of the field, and the [AI coding agents hub](/ai-coding-agents/) and [pricing pillar](/ai-pricing/) lay out the full ladder across tools for a side-by-side view.

## Frequently asked questions

How much does Kiro cost?

Kiro Free is $0/mo for 50 credits, permanently. Pro is $20/mo for 1,000 credits, Pro+ is $40/mo for 2,000, Pro Max is $100/mo for 5,000, and Power is $200/mo for 10,000\. Every paid tier costs the same $0.02 per credit; only the total pool size changes.

What is a Kiro credit?

A credit is Kiro's unit of usage, consumed fractionally in 0.01 increments based on how complex a request is. Kiro does not publish a fixed credits-per-task table; simple edits use less, long or complex agent runs use more.

Does Kiro have a free tier?

Yes. Kiro Free includes 50 credits every month, permanently, with access to open-weight models and Claude Sonnet 4.5 under rate limits. It does not expire and does not require a paid card.

Does using Auto or picking a specific model change what Kiro costs?

Yes. Kiro's default Auto agent costs 1x credits per task. Pinning Claude Sonnet by name instead of using Auto costs 1.3x credits for comparable work, roughly a 23% credit premium for hand-picking the model.

What happens if you run out of Kiro credits?

By default, nothing bills automatically: the agent stops working once the pool is spent. Paid tiers can opt in to overage credits at $0.04 each, charged at the end of the billing cycle, but only if that setting is turned on first.

## Sources

* Kiro (Cognition / AWS). (2026). _Pricing_ \[vendor pricing page; plan tiers, credit allotments, overage rate, sign-up bonus\]. [kiro.dev/pricing](https://kiro.dev/pricing/)
* Kiro (Cognition / AWS). (2026). _Billing for individuals_ \[vendor documentation; fractional credit consumption, opt-in overage mechanics\]. [kiro.dev/docs/billing](https://kiro.dev/docs/billing/)
* Kiro (Cognition / AWS). (2025). _Announcing new pricing plans and Auto, our new agent_ \[company announcement, September 15, 2025; unified credit pool, Auto’s 1x versus pinned-Sonnet 1.3x credit cost\]. [kiro.dev/blog/new-pricing-plans-and-auto](https://kiro.dev/blog/new-pricing-plans-and-auto/)