AI coding assistant pricing: seats, credits and limits

Compare the work and billing unit first

You want help shipping changes, but the plan selector offers seats, credits, usage pools and higher limits. Start with three questions: Do you mostly accept inline suggestions, ask an agent to change several files, or run terminal tasks? Which models are required? What happens when the included allowance runs out? A lower subscription price answers none of these on its own.

For a first comparison, write down your editor, repository language, required model and monthly cash ceiling. Include one person who reviews generated changes. Then separate the access fee from additional purchases and from the engineering time needed to make a change usable. Our recommendations below describe documented workflow fit; we did not run a coding-quality or speed benchmark.

Use a short trial to measure your own workload. A credit is a vendor-defined unit. One Kiro credit, one GitHub AI credit and a Cursor usage allowance cannot be converted into the same number of completed fixes from their names. A plan with more credits is not automatically the plan with more useful work.

Cursor, Copilot, Claude Code and Kiro compared

Cursor: the current page lists Plus at $8/month, restricted to Grok models, and Premium at $30/month. Premium lists cloud agents and automations; third-party model use is charged separately at API list prices. Choose the lower tier only if its model restriction fits your work. Avoid selecting it on price alone when your task requires another model. The displayed Cursor/Grok allowance renews weekly, which matters if most of your work happens in a few concentrated days.

GitHub Copilot: individual Pro is $10/month, Pro+ $39 and Max $100. Paid plans list unlimited completions and next-edit suggestions; chat, agents, code review and CLI work use AI credits. Choose Pro as a candidate when inline assistance is the main job and you can track the separate agent allowance. Avoid assuming unlimited completion means unlimited agent execution. The plans page includes variable flex credits alongside the base allowance; do not build a long-term budget around a temporary bonus.

Claude Code: Claude Pro includes access at $20 on monthly billing, or $200 paid for a year. Max tiers are listed at $100 and $200/month. Choose this route when you want Claude Code together with Claude’s paid application access and can work within its usage limits. Avoid treating a subscription as an unlimited API account: the API is a separate purchase route. The annual price is a commitment, not twelve independent $17 payments.

Kiro: Pro is $20/user/month with 1,000 credits, Pro+ $40 with 2,000, Pro Max $100 with 5,000 and Power $200 with 10,000. Choose a tier after measuring credit use on a representative task. Avoid buying capacity from an assumed prompts-per-credit ratio. Complexity affects consumption, so a small edit and a repository-wide change do not have a fixed equal cost.

Worked example: the $80 team budget that becomes $100

We performed this editorial calculation in Node.js on 15 October 2026. Fictional inputs: four separately billed individual Kiro Pro subscriptions, 1,100 credits used per person in one month, no existing add-on balance, and a $90 cash budget. This is a hypothetical workload, not a measurement of how many credits a real coding task consumes.

Initial worksheet: four subscriptions × $20 = $80, apparently within budget. It omitted extra capacity. At 1,100 credits per person, each person needs 100 beyond the 1,000 included. The current individual-plan add-on system is prepaid, with a minimum $5 purchase for 125 credits. Buying one minimum pack per person adds $20 to the cash bill. Final bill: $100, which exceeds the budget by $10. Each person retains 25 add-on credits; their expiry still matters. The consumed extra capacity has a $16 nominal value, but $96 is not the cash payment in this scenario.

We also corrected a separate annual-billing error: multiplying Claude’s rounded $17 monthly equivalent by twelve gives $204; its stated annual upfront charge is $200. Our script deliberately retained both incorrect initial values, rejected them against the expected results, then passed four checks on the corrected calculation, including the budget decision and a below-allowance case.

Run the downloadable file with “node coding-budget.mjs”. To adapt the exercise for planning, change the inputs and the expected checks together. The result excludes taxes, other subscriptions, hosting and developer time. Enterprise Kiro overages follow different terms; this example applies only to the stated individual subscriptions and chosen pack purchases.

Why a credit balance does not predict completed tasks

GitHub documents one AI credit as $0.01 of usage value. That is a billing conversion, not a promise that every request costs one credit. Model selection, context and the work performed affect consumption. Use the billing model that actually applies to your account; older annual Copilot subscriptions may still have legacy request-based terms.

Cursor’s third-party model charges are separate from its subscription display. If you require those models, record that additional usage as its own budget line. Do not compare a subscription-only number for one tool with a subscription-plus-usage total for another.

Claude usage is shared across its application surfaces and Claude Code. Its pricing materials describe rolling five-hour limits and additional paid-plan weekly limits. A quiet week followed by an intensive migration can feel different from an even workload. We have not established a reliable number of completed tasks for any tier. Check your account’s current limits before scheduling work that cannot pause.

Run one repository task before upgrading

Pick a small change with an observable outcome: for example, fix a date filter that excludes the final day of a report. Save the starting commit, the expected behavior and an existing failing test. Give each candidate the same task description and scope. Record the plan, model, date and starting usage balance. This is a proposed product evaluation; we did not run it in the four vendor applications.

Review the diff before accepting it. Run the relevant tests, inspect boundary cases and count manual corrections. Record accepted changes, unresolved defects, elapsed review time and the usage charged. If a tool changes unrelated files or bypasses the intended constraint, include that correction work in your assessment. A convincing explanation is not evidence that the change works.

Repeat with one task that resembles your normal week rather than extrapolating a trivial edit to a large migration. Use a permitted sample repository. Your final comparison should say which workflow you tested and what remains unknown. It should not turn a single completed task into a universal ranking.

Choose monthly, annual or team billing deliberately

Monthly billing is useful while the workload and tool fit remain uncertain. An annual commitment can reduce the stated equivalent cost, but the useful comparison is the amount due now and the period purchased. In the Claude example, $20 monthly totals $240 over twelve unchanged renewals; $200 buys a year upfront. That $40 difference does not establish the value of a plan you may stop using.

Do not multiply individual offers into a team procurement quote without checking the team product. Cursor advertises Teams from $30/user/month, while organizational Copilot billing has its own allowances and pooling rules. Central administration, identity controls and shared budgets may change which product is appropriate. We did not verify negotiated enterprise discounts, every regional tax treatment or a complete security-control comparison.

Keep the model bill, developer time and downstream services visible. If a tool generates an app, its deployment and database bills belong in the project budget as well. A subscription comparison answers what access costs; your trial establishes whether the resulting changes are worth reviewing.

Make the purchase decision from a short ledger

Keep a ledger with the plan name, billing interval, number of seats, allowed models, included allowance, extra-purchase rule, spending control and next reset date. Add the date of the source check. Leave a field unknown when it is unknown; do not turn an unverified feature into a reason to upgrade.

Choose the smallest plan that passes your actual workflow and budget checks. Revisit it when the model, repository size or number of users changes. If you repeatedly hit a limit, compare the next tier with the real additional purchases you made, including unused prepaid balances. Keep the accepted-change evidence beside the bill so you can distinguish a capacity problem from a quality or review problem.

Continue with the right comparison

Use the coding-assistant hub to widen the shortlist, or a tool profile to inspect its recorded capabilities. This guide is the dated billing comparison. When you need an app generated from a brief with accounts, persistent data and a developer handoff, use the app-builder category rather than assuming every coding subscription offers that complete workflow.

Profiles and linked comparisons may be checked on different dates. For a purchase, use the official pricing links attached to this guide and confirm the terms shown to your account at checkout.

Coding assistants questions