Core knowledge · Chapter 6 of 21
Credits & Pricing
How Kiro's credit system actually works: what a credit buys, how model multipliers change the bill, and how to avoid surprise overages.
All levels last reviewed 2026-08-08
◎ Learning objective
Explain what a credit is, predict roughly what a task will cost on a given plan and model, and name the habits that prevent surprise bills.
Most people meet Kiro’s pricing page, see five plans and a credit number, and assume credits work like API tokens. They don’t. Credits measure work done, not text processed, and the same piece of work can cost more than double or a twentieth as much depending on one dropdown setting. This page explains the model well enough that you can estimate a month before it happens.
What a credit actually is
Simple definition. A credit Kiro's unit of billing: one unit of agent work performed in response to a prompt, billed fractionally. is a unit of work the agent does in response to a prompt. Not a message, not a token, not a minute. A short question (“what does this function return?”) costs less than one credit. A big piece of work, like executing one task from a spec’s tasks.md, typically costs more than one.
Analogy. A taxi meter, not a bus ticket. A bus ticket charges the same whatever the ride; a meter charges for distance and time. Kiro’s meter runs on how much work your request causes: how many files the agent reads, how many edits it makes, how many commands it runs and re-checks.
Three mechanical facts worth knowing:
- Billing is fractional, in increments of 0.01 credits. You are not rounded up to a whole credit for asking a one-line question.
- One unified pool. Since October 2025 every request type draws from the same credit balance. Older articles describe separate vibe-request and spec-request limits; that split is gone.
- Autocomplete isn’t on the meter. Kiro’s official FAQ defines billable work as prompts you ask Kiro to execute (vibe or spec), spec refinement, task execution, and agent hook execution. Inline suggestions as you type aren’t on that list; the meter runs on agentic requests.
In short: credit = one unit of agent work; fractional; one shared pool; autocomplete is free.
The five plans
Prices are per user, per month, verified against kiro.dev/pricing on the review date at the bottom of this page.
| Plan | Price | Credits / month | What it unlocks |
|---|---|---|---|
| Free | $0 | 50 | Sonnet 4.5 + open-weight models only |
| Pro | $20 | 1,000 | All premium models, Kiro Web, overage allowed |
| Pro+ | $40 | 2,000 | Same, with headroom for daily professional use |
| Pro Max | $100 | 5,000 | Added June 2026; for heavy or premium-model use |
| Power | $200 | 10,000 | Highest individual ceiling |
A few things the table can’t show:
- Going over costs $0.04 per credit either way: individual plans top up with add-on packs, and team plans can let admins enable automatic overage. The Free tier simply stops at 50 until the month resets.
- Add-on credit packs run from $5 (125 credits) up to $100, for when you’d rather top up than change plans. Unlike the monthly allowance, pack credits stay valid for 12 months.
- Kiro Web (app.kiro.dev) requires a paid subscription. The Free tier is IDE and CLI only.
- Team plans mirror the individual prices and add consolidated billing, usage analytics, SAML/SCIM single sign-on through AWS IAM Identity Center, and organization management. Same credits, different administration.
Notice that every paid tier prices credits identically: $20 ÷ 1,000, $40 ÷ 2,000, $100 ÷ 5,000, $200 ÷ 10,000 all come to $0.02 per credit. That single number is what makes the overage rule below so easy to reason about.
Model multipliers: the part everyone misses
Your credit balance drains at a rate set by the model you have selected. Every model carries a multiplier A per-model factor applied to credit consumption, expressed relative to Auto at 1.0x. relative to Auto, which is defined as 1.0x.
Take one moderately involved task — say, implementing a spec task that touches four files and runs the test suite twice — and assume it costs 10 credits on Auto. The same task, same result, different model:
| Model | Multiplier | That same task costs |
|---|---|---|
| Qwen3 Coder Next | 0.05x | ~0.5 credits |
| Claude Haiku 4.5 | 0.4x | ~4 credits |
| Auto (router) | 1.0x | 10 credits |
| Claude Sonnet 5 | 1.3x | ~13 credits |
| Claude Opus 5 | 2.2x | ~22 credits |
| GPT-5.6 Sol | 2.4x | ~24 credits |
That is a 48x spread between the cheapest and the most expensive option for identical work. No other habit you can change comes close to mattering this much. The full multiplier table for every available model lives in Choosing a Model.
In short: the model dropdown is the main dial on your monthly bill; Auto is the sane default.
Bonuses when you start
Two one-time sweeteners, both easy to miss:
- New users get 500 bonus credits, valid for 14 days. That’s ten months of Free-tier allowance, expiring in two weeks — so use it on something real. A full spec-driven feature, not a tour.
- The first upgrade from Free applies a $20 credit to your first paid month. Pro’s first month effectively costs nothing.
Three rules that bite
Credits do not roll over. An unused balance disappears at the monthly reset. Nobody banks credits for a big month later; if you finish a month at 200 of 1,000, those 800 are gone. Plan for your typical month, not your worst one.
Overage is exactly double the tier price. Every paid tier sells credits at $0.02; overage sells them at $0.04. Concretely: if you’re on Pro ($20 for 1,000) and consistently use 2,000, you pay $20 plus 1,000 × $0.04 = $60 a month. Pro+ gives you the same 2,000 credits for $40. The add-on packs price the same way — $5 for 125 credits is also $0.04 per credit.
Subscriptions are per individual. A plan belongs to one person; you can’t share a Power seat across a team. That’s what the team plans and their consolidated billing are for.
Which plan fits you?
The daily-burn figures below are planning heuristics for sizing a plan, not published figures — your real usage depends on how large your tasks are. Do the arithmetic with your own numbers after a week of real work.
The evening hobbyist. A few sessions a week, small projects, mostly Auto. Call it 5–10 credits per evening session, three evenings a week: roughly 60–120 credits a month. Free (50) runs out just as things get interesting; Pro at 1,000 gives you a ceiling you’ll never think about. Start Free, upgrade when you hit the wall.
The full-time developer. Kiro open all day on Auto and Sonnet 5, running specs and reviewing diffs: roughly 30–50 credits a day. Over ~21 working days that’s 630–1,050 credits a month — landing right on Pro’s 1,000 ceiling, which means the good months fit and the busy months spill into overage. Pro+ (2,000) is the comfortable resting place; a month heavy on spec execution can justify Pro Max.
The premium-model user. Same working day, but Opus 5 selected throughout. Multiply everything by 2.2: 66–110 credits a day, 1,400–2,300 a month. That blows past Pro+ in any busy month, so Pro Max (5,000) is the realistic floor. Add parallel agent sessions in Agent Focus mode, or a habit of running GPT-5.6 Sol at 2.4x, and you’re in Power (10,000) territory.
Read those three together and the pattern is clear: hours worked moves your bill linearly, but model choice moves it by more than a factor of two — in either direction.
Four habits that prevent surprise bills
- Default to Auto; escalate deliberately. Let the router pick. Reach for Opus 5 or GPT-5.6 Sol when you have a genuinely hard problem, and switch back afterwards. Treat “which model is selected right now?” as a question worth asking before a long session.
- Check usage weekly, not monthly. Multiply your first week by four. If the answer exceeds your plan, you have three weeks to adjust rather than a bill to explain.
- Watch anything running unattended. Headless CLI runs and CI automation spend credits with nobody at the keyboard, and a retry loop that would annoy you interactively will quietly bill you at 3am. Scope them tightly and cap them. See Kiro CLI Guide for the headless flags.
- Front-load thinking into specs. Specs cost credits to produce, but a spec that prevents one round of “no, rebuild that differently” pays for itself several times over. Vague prompts are expensive twice: once for the wrong answer, once for the redo.
A caveat about all of these numbers
Pricing changes. Kiro added a whole tier (Pro Max) in June 2026 and gained a second model vendor in July 2026, each of which shifted the arithmetic on this page. The multipliers move as models are released and retired.
kiro.dev/pricing is the authority, always. Everything here was verified against it on the last-reviewed date shown on this page. If the date is more than a couple of months old and you’re about to make a spending decision, check the source first.
☰ Chapter summary
- Kiro bills in credits: Free $0/50, Pro $20/1,000, Pro+ $40/2,000, Pro Max $100/5,000, Power $200/10,000 per user per month.
- A credit is a unit of agent work, billed fractionally in 0.01 steps; simple asks cost under 1 credit, executing a spec task costs more.
- Every request draws from one unified credit pool (since October 2025); billable work is agentic requests, not inline autocomplete.
- Model choice is the biggest lever: the same task costs 2.2x more on Opus 5 than on Auto, and 0.4x on Haiku 4.5.
- Credits do not roll over, and overage at $0.04/credit is exactly double the $0.02/credit every paid tier charges.
- Free tier is Sonnet 4.5 plus open-weight models only; premium models and Kiro Web need a paid plan.
All chapter summaries are collected on the revision page.