Kiro CLI vs Claude Code: pricing, licenses and the cost path

Kiro CLI and Claude Code are both current, first-party, closed-source coding agents — but they are not symmetric on the thing most people open this comparison to settle. Kiro CLI's cost path is a closed AWS subscription metered in credits, with no documented way to point it at a model endpoint of your own. Claude Code's is either an Anthropic subscription or per-token billing to whatever Anthropic-format endpoint you set ANTHROPIC_BASE_URL to. If you are comparing because you want to own the cost path, that is a property of Claude Code and not of Kiro CLI, and no amount of configuration changes that.

One naming fact before anything else, because it decides which prices are even relevant. Kiro CLI is the renamed Amazon Q Developer CLI. And "Kiro" is one letter from "Kilo Code", a different product that does accept custom OpenAI-compatible endpoints — see Kilo Code vs Claude Code if that is the tool you meant. Every sentence about custom endpoints below names its product for that reason.

What you are actually comparing

Kiro's migration guide gives the dates: "Kiro CLI will be available starting November 17th, 2025", and machines with auto-update enabled were moved on November 24th, 2025. The old entry points survive — "q and q chat will continue to work, although we recommend using kiro-cli" — and installation copies prompts and agents from ~/.aws/amazonq into ~/.kiro, with rules landing in ~/.kiro/steering. The same page records a change that no plan table shows: "the Kiro CLI software is licensed under the AWS Intellectual Property License, while the Q Developer CLI software was licensed under Apache 2.0." The rename took the tool out of open source. AWS states the rebrand plainly in its own upgrade doc.

That lineage is also why the predecessor keeps showing up in results. AWS's end-of-support notice says: "On April 30, 2027, AWS will discontinue support for Amazon Q Developer IDE plugins. For capabilities similar to Amazon Q Developer IDE plugins, explore Kiro", and the AWS DevOps blog post of April 30, 2026 adds that new signups stopped on May 15, 2026.

A conflict worth catching before you budget. As of September 19, 2026, aws.amazon.com/q/developer/pricing/ still advertises Amazon Q Developer Pro at $19 per user per month with a free tier and carries no sunset banner, while the docs and blog above say signups closed and support ends in 2027. AWS's marketing page and AWS's documentation disagree. Neither number is Kiro CLI's price; kiro.dev/pricing is, and it is denominated in credits. Confirm in checkout rather than from either page.

Versions, so nobody installs the wrong line. The Kiro CLI changelog's newest entry is 2.22.0, dated September 16, 2026; a v3 line exists as opt-in early access launched with kiro-cli --v3 alongside an existing 2.x install, so 2.x is the shipping default. Claude Code ships as the npm package @anthropic-ai/claude-code, latest 2.1.278 on the npm registry, with a license field of "SEE LICENSE IN README.md" — proprietary, like Kiro CLI now is. Anthropic's Claude Code documentation now serves from code.claude.com, and the links below use that host rather than the older docs.claude.com paths.

Claude Code vs Kiro CLI: who should pick which

Five decisions actually separate these two. None of them is "which model is better" — nothing on this page was runtime tested, and neither vendor publishes a head-to-head benchmark worth quoting.

Pick Kiro CLI if your account is already the deciding factor. If you are on Amazon Q Developer today, the migration is done for you and the predecessor is on a clock. If your organization signs in through AWS IAM Identity Center or an external IdP, Kiro accepts those directly. And one subscription spans the surfaces: kiro.dev/pricing says subscriptions "can be used with Kiro IDE, Kiro CLI, Kiro on the web, Kiro Crew, ACP compatible IDEs, and automation in software development." If you want the agent driven from JetBrains or Zed rather than a terminal, kiro-cli acp exposes it over ACP, which Kiro describes as "an open standard that enables AI agents to work with any compatible editor" and documents against JetBrains IDEs, Zed and other ACP-compatible editors.

Pick Kiro CLI if you want the model list locked by an administrator. Kiro's enterprise governance post (March 12, 2026) says disabled models "don't appear in the model selector for any developer in the organization for both the IDE and the CLI." That is a real control, and it is the same property that closes the endpoint question below.

Pick Claude Code if you want per-token accounting. A Kiro credit is defined by Kiro's own FAQ as "a unit of work in response to user prompts", where simple prompts can consume less than one and a spec task typically more. That is honest and it is also unforecastable: you cannot tell in advance what a month of your work costs in credits, and you cannot convert credits into tokens because no conversion is published. Per-token billing gives you a rate you can multiply.

Pick Claude Code if the cost path itself is the decision. This is the asymmetry. Claude Code documents an endpoint override; Kiro CLI documents the opposite of one.

Pick on quota shape if you keep hitting walls. Kiro publishes monthly credit quotas that reset at the start of the billing month, and its FAQ says unused subscription credits do not roll over. Claude Code on a Team or Enterprise plan draws from "a per-seat allowance that resets on a rolling five-hour window and a weekly window", per Anthropic's cost documentation, which adds that a session or weekly limit is "shared across all models, so the developer can't restore access by switching models with /model." These fail differently: one runs out for the month, the other pauses you for hours. Kiro publishes no session-window equivalent that this page found, which is not the same as Kiro having none.

The published differences, side by side

Kiro CLI pricing and Claude Code pricing, in their own units

Read from kiro.dev/pricing and kiro.dev/faq on September 19, 2026. The only arithmetic those two rates support is the overage ratio: Pro's included rate works out to $0.02 per credit, so an add-on credit costs twice what an included one does. That says nothing about work done, because a credit is not a fixed amount of work.

Read from claude.com/pricing on September 19, 2026. Two honest limits on that table. The Free row simply does not list Claude Code among its features — that is an absence observed on the page, not a quoted exclusion sentence. And the Max row shows the "From $100" floor with 5x and 20x usage levels; this page did not capture a first-party price string for the higher level, so no second figure is printed. Claude plan limits covers how those allowances behave in practice, and Claude vs Claude Code covers the fact that one subscription serves the chat apps and the CLI from the same allowance.

Why Kiro credits and token rates cannot be compared directly

Kiro publishes a per-model credit multiplier, which looks like a conversion table and is not one — the multiplier scales credits, and a credit is still a unit of work rather than a token count. Selected rows:

Read from kiro.dev/docs/models/available-models/ on September 19, 2026. Several models in that list — including Claude Sonnet 5, the Opus 4.x entries and Claude Haiku 4.5 — carry no published multiplier at all, and this page does not interpolate one from a neighbouring model. The free tier is capped further: kiro.dev/pricing footnotes that free-tier users who signed up with social logins or AWS Builder ID reach Claude Sonnet 4.5 and open-weight models such as Qwen3 Coder Next, DeepSeek 3.2 and MiniMax M2.1 — so the Opus line is not in reach at $0.

Anthropic, by contrast, publishes the one first-party per-developer cost anchor either product has. Its cost documentation states that across enterprise deployments "the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users." That is Anthropic's own figure for token-billed Claude Code, not a Kunavo measurement and not a quote for your team.

The cost path: what Claude Code lets you change, and what it costs you

Anthropic documents the endpoint override and its consequences on the same pages, which makes it checkable in both directions. The configuration is two variables, and the base URL is an origin: Anthropic's gateway compatibility guide lists the endpoints an ANTHROPIC_BASE_URL gateway serves as /v1/messages and the optional /v1/messages/count_tokens, and its own examples set the variable to a bare host, so do not put /v1 in the value yourself.

# ANTHROPIC_BASE_URL is an ORIGIN. Claude Code appends the route itself.
export ANTHROPIC_BASE_URL=https://api.kunavo.com
export ANTHROPIC_AUTH_TOKEN=sk-kn-...

# Anthropic's own verification call, which shows the concatenation:
curl -X POST "$ANTHROPIC_BASE_URL/v1/messages" \
  -H "Authorization: Bearer $ANTHROPIC_AUTH_TOKEN" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "claude-sonnet-4-6", "max_tokens": 1, "messages": [{"role": "user", "content": "."}]}'

Three things are worth reading carefully before you treat this as free money. First, the credential is what moves the billing, not the base URL. Anthropic's gateway overview says that while a gateway credential variable or apiKeyHelper is active, "a developer's claude.ai subscription isn't used" and its usage limits do not apply — and that "setting only that variable, without a gateway credential, doesn't replace the subscription."

Second, features drop off. The connect page states that Remote Control and voice dictation "are unavailable while ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN, or an apiKeyHelper is active", that Remote Control is additionally disabled while the base URL points at a non-Anthropic host, that the fast-mode availability check "goes directly to api.anthropic.com and doesn't follow ANTHROPIC_BASE_URL", and that with a credential variable also active Claude Code "doesn't report usage metrics to the Console analytics dashboard". In the desktop app, gateway configuration also removes SSH and Anthropic-hosted cloud environments from the environment picker. The agent loop itself still runs over the Messages API, but "unaffected" would overstate it: Anthropic's gateway compatibility guide lists capabilities that depend on the gateway forwarding the right headers and body fields — context management, beta tool fields, prompt-cache markers, token counting — and records that fine-grained tool streaming is off by default on any custom base URL. What a given gateway preserves is a property of that gateway, not of Claude Code.

Third, Anthropic bounds what anyone may claim about this. The same overview says: "Any gateway that exposes a supported API format works. Anthropic doesn't endorse, maintain, or audit third-party gateway products, and doesn't support routing Claude Code to non-Claude models through any gateway." Kunavo serving Claude models over an Anthropic-format endpoint sits inside that documented shape. Running a non-Claude model inside Claude Code does not, and this page does not suggest it.

Kiro CLI has no equivalent section, and that is the finding rather than a gap in the research. Its authentication page documents GitHub, Google, AWS Builder ID, IAM Identity Center and external IdP sign-in and no third-party provider key, and states that "API key based authentication is only available for Kiro Pro, Pro+, Pro Max, and Power subscribers." Its headless documentation covers --no-interactive, --trust-all-tools, --trust-tools, --agent-engine, --output-format stream-json and --require-mcp-startup, with no base-URL or endpoint option anywhere. Bring-your-own-key is an unmet community request — the flagship issue was opened on June 11, 2026 and closed as a duplicate with no maintainer commitment. So there is no documented Kiro CLI endpoint setting to show here. That is a statement about Kiro's published surface, not proof that nothing anywhere can be made to work — but a config block presented as an official Kiro setting is not one.

A worked estimate for the Claude Code route

This is illustrative token arithmetic, not a measured task cost and not a bill ceiling. Assume one bounded task sends 250,000 uncached input tokens and receives 15,000 output tokens, and that a working day contains 8 such tasks. Those three numbers are assumptions for illustration; your repository decides the real ones. Rates are live Kunavo catalog prices per million tokens.

Read it as a shape rather than a quote. Under these assumptions Claude Sonnet 4.6 models $3.12 for the assumed day against $5.20 on Claude Opus 5 — which is why model selection moves a token bill more than provider selection does, and why Anthropic's own guidance is to reserve the heaviest model for the work that needs it. Prompt caching changes these numbers substantially in real sessions and is excluded here; see Claude prompt caching. Cheapest listed rate and cheapest way to finish the task are also different questions: a model that needs three attempts can cost more than one that needs one.

Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. The minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee and not a subscription. See billing details.

Migration cost, in both directions

Moving into Kiro CLI from Amazon Q Developer is the cheapest migration on this page, because AWS did it for you — the file copy described above happens at install, and the old commands keep working. The cost is the license change, which is a procurement question rather than an engineering one.

Changing Claude Code's cost path is two environment variables and no reinstall, and it is reversible in the same breath: unset them and the saved claude.ai login, which has been sitting unused on disk, takes over again. That reversibility is the practical reason to try the endpoint route on one repository before committing a team to it. What does not migrate in either direction is agent and prompt configuration — the two products keep their files in different directories, and neither documents reading the other's.

If the cost path is your reason for comparing

Then the decision is already made by the products rather than by preference: Kiro CLI has no documented endpoint setting, and Claude Code does. Kunavo publishes a setup guide for the Claude Code side — a configuration reference, not a compatibility test; Claude Code has not been runtime-tested here against Kunavo's endpoint, and nothing on this page claims it has. Keep a working route available while you try it, run one bounded task, then read the charge your account actually recorded for it. Start at the Claude Code integration guide, and create a Kunavo account when you are ready to fund a key.

Still narrowing the field? Claude Code pricing breaks down the subscription-versus-token decision on its own, Claude Code without a subscription covers the endpoint route end to end, and Copilot CLI vs Claude Code plus OpenCode vs Claude Code compare two more terminal agents with different billing shapes.

FAQ

Every official source linked on this page was read on September 18 or 19, 2026: kiro.dev pricing, FAQ, changelog, authentication, headless, ACP, available-models and governance pages; the Kiro migration guide; AWS's end-of-support notice, upgrade doc, DevOps blog post and the Amazon Q Developer pricing page; the Kiro issue tracker; claude.com/pricing; Anthropic's gateway, gateway-connect and cost documentation; and the npm registry. Neither client was runtime tested against any endpoint for this page, no benchmark or performance comparison is made, and every dollar example is illustrative token arithmetic at live Kunavo catalog rates rather than a measured task cost.