Kiro

A credit is Kiro’s unit of work per prompt, not a token count. The billing questions page says a simple prompt can use less than one credit, a spec task typically uses more, and usage is metered to the second decimal place. The model you pick multiplies the cost. The models page sets Auto at 1.0x and gives the example that a task costing 10 credits on Auto costs 20 on Claude Opus 5.5 and 0.5 on Qwen3 Coder Next.

Two things make a task cost more than expected. Equal multipliers do not mean equal spend, because models generate different amounts of output and reasoning. And a request with very large context or output can count as more than one credit. Kiro shows the count right after each interaction and updates the account total at least every five minutes.

A spec turns a feature request into three files in .kiro/specs/: requirements.md with user stories and acceptance criteria, design.md with the architecture, and tasks.md with a trackable implementation plan. You can start from requirements or from the technical design; bugs get a bugfix.md instead. Each spec task the agent executes uses credits like any other prompt.

Steering files are Markdown documents that give the agent your standards; workspace files sit in the project, global ones in your home folder, and teams can push global files to machines through device management. Hooks are JSON files that run a shell command or an agent prompt when an event fires, such as the agent saving a TypeScript file. A hook that runs an agent prompt also uses credits each time it fires.

The billing page separates the first upgrade from later ones. Your first move from Free to a paid plan is prorated to the days left in the month but gives the full credit allowance at once. After that first month, an upgrade charges the whole price difference and re-rates the month under the bigger allowance, while a downgrade charges the current plan in full and takes effect next month.

Cancelling means switching to Free: the cancellation guide keeps the paid plan to the end of the month and leaves unused add-on credits valid until they expire. The FAQ adds that a Kiro subscription works in the IDE, CLI, Web, Kiro Crew, ACP-compatible editors and CI automation, but not through third-party harnesses that route requests outside Kiro’s own interfaces.

It depends on how you sign in. The data protection page says content from Free and individual subscribers, meaning paid users who sign in with GitHub, Google or AWS Builder ID, may be used for service improvement unless they opt out. That content includes questions, other inputs, and the responses and code Kiro generates. Free inputs may also be kept up to 60 days for abuse detection. For both groups content is stored in the US East (N. Virginia) Region, while models may process requests elsewhere.

Enterprise users signing in through IAM Identity Center or their own identity provider are opted out automatically. Everyone else can turn off usage analytics and content collection separately in the IDE under Settings, Application, Telemetry and Content, or in the CLI settings.

AWS has set the dates. Its end-of-support post of 30 April 2026 says Q Developer IDE plugins and paid subscriptions reach end of support on 30 April 2027, new Q Developer sign-ups stopped on 15 May 2026, and the latest coding models are available only on Kiro. Existing Pro subscriptions can still add users, and Java upgrades and .NET porting through the plugins end on the same date. Q Developer inside the AWS console and chat apps is not affected.

Kiro is a separate subscription, priced per user in credits rather than the Q Developer Pro seat. If you would rather stay in a VS Code-style editor with a different usage model, the Devin Desktop profile covers the former Windsurf IDE.

1. Install the IDE, sign in with GitHub or Google, and turn off content collection under Telemetry and Content if you want to keep prompts out of service improvement.
2. Open a small repository and add one steering file with your test command and code style.
3. Ask for a spec for one small feature on Auto, and check requirements.md and design.md before letting tasks run.
4. Run the tasks one by one and note the credit count Kiro shows after each, plus the total on the usage dashboard.
5. Multiply the credits per task by your expected monthly tasks and by the multiplier of the model you would use.

Accept if the tasks pass your tests and the projected month fits a plan’s allowance. Reject if the spec adds requirements you did not ask for, or one feature uses a large share of the 50 free credits. No result is recorded here.