If you want predictable Replit Agent spending, set a service shutdown limit before choosing a faster or more capable mode. You cannot calculate every request in advance from a fixed price list: Agent charges for the work it performs, then deducts that usage from your Replit credits. This is Replit Agent effort-based pricing explained as of 2026-07-20.
How effort-based charging works
Replit’s AI billing documentation says all Agent interactions are billable, including text-only answers and Plan Mode conversations. In Build Mode, Agent normally creates one checkpoint when it completes a request that changes code. A small fix typically costs less than a complex feature or integration, while a larger request is bundled into one checkpoint reflecting the total effort. No checkpoint does not necessarily mean no charge.
There is therefore no documented flat dollar price for each prompt. Monthly plan credits cover Agent and other Replit services, including publishing, storage, and databases. Usage beyond those credits can become an additional charge. Checkpoint costs are visible from the Agent tab by hovering over the usage icon; the Account usage page shows the current billing-period breakdown. Replit warns that dashboard data can take up to 30 minutes to appear, as of 2026-07-20.
Lite, Economy, Power, and Turbo
The names describe cost/capability controls, not prepaid bundles. According to the current Agent modes documentation, Lite, Economy, and Power are the three base modes. Turbo is an optional Power-mode toggle.
| Choice | What Replit documents | Cost implication as of 2026-07-20 |
|---|---|---|
| Lite | Lightweight models for quick, scoped edits | Focused requests often cost less than the same edit in Economy or Power |
| Economy | Cost-optimized mode for everyday builds | Uses fewer credits per task |
| Power | More capable models for complex work and larger codebases | Intended for harder work rather than minimum spend |
| Power + Turbo | Fastest available models; Pro and Enterprise only | About 2x the cost of Power, with responses advertised as 2.5x faster |
High effort is another optional toggle for Economy and Power, not a separate tier. Replit says it can add up to approximately 2x the cost on the toughest tasks, while adding little or nothing on simpler ones, as of 2026-07-20. Complex features, large refactors, integrations, extended reasoning, Turbo, and paid third-party APIs used by Agent can all raise consumption. Those API services are billed at the provider’s public API rate and deducted from Replit credits.
Set an alert and a hard cap
For a personal Core account, follow the path in Replit’s billing overview and spend-management guide:
- Select your profile icon in the top-right corner.
- Select Settings.
- Under Account, select Billing.
- Set the usage alert threshold for an email warning.
- Set the service shutdown limit, the hard cap for usage-based services beyond included credits.
The alert is a warning; it does not stop Agent. The service shutdown limit is the hard control: when reached, usage-based services are suspended until you raise the limit or the next billing cycle begins. Because the control applies across usage-based services, it is not an Agent-only ceiling.
For an organization, open its usage page, select Manage under Usage total, and set the usage limit. You need organization admin or owner permission. Enterprise admins can also edit a member’s limit from the Agent users table.
One important ordering detail: Replit says limits apply after purchased credit packs are consumed. If you want usage to stop after the pack rather than continue into pay-as-you-go charges, its official guide says to set the limit to $0.01 as of 2026-07-20.
A practical default
Use Lite for a precisely scoped edit, Economy for cost-conscious everyday work, and Power only when complexity warrants it. Leave Turbo and High effort off until the task justifies their documented cost tradeoff. Then use the checkpoint cost and Account usage breakdown to calibrate future requests, but rely on the service shutdown limit, not a delayed dashboard reading, to enforce your maximum.
Who needs which control
If you use Agent for occasional, precisely scoped edits, mode selection is your first cost lever. Lite is aimed at quick scoped work, while Economy is the cost-optimized everyday option. You can start there and inspect the resulting usage rather than choosing Power for work that does not require its more capable models.
If you build larger features, integrations, or refactors, the uncertainty comes from effort rather than a fixed prompt price. Power may fit the complexity, but Turbo and High effort can raise consumption further. The useful comparison is the same task with the least expensive mode that completes it acceptably, followed by the checkpoint and account-usage record.
If you administer a team, the hard control matters more than any individual’s mode preference. Organization administrators or owners can set the usage limit, and Enterprise administrators can edit a member’s limit from the Agent users table. Because the service shutdown limit applies across usage-based services, you also need to account for publishing, storage, databases, and paid third-party APIs that draw from the same credits.
Before you start a costly request
- Check the remaining plan credits and current billing-period usage. Do not infer the balance from the number of checkpoints.
- Set the service shutdown limit first. An email alert warns you; it does not suspend Agent or other usage-based services.
- Choose the lowest mode that matches the task. Use Lite for a focused edit, Economy for cost-conscious everyday work, and Power when the complexity warrants it.
- Leave Turbo and High effort off unless you need their tradeoff. Turbo is about 2x Power’s cost, while High effort can add up to approximately 2x on the toughest work and may add little or nothing on simpler tasks.
- Check for paid third-party APIs in the request. Their public API rates are deducted from Replit credits alongside Agent usage.
- Inspect both cost views after the run. Hover over the usage icon for the checkpoint cost and use the Account usage page for the billing-period breakdown. Allow for the documented delay of up to 30 minutes.
- If you bought a credit pack, confirm the stopping behavior you want. Limits apply after packs are consumed; the guide’s stated setting for stopping after the pack is $0.01 as of 2026-07-20.
Common pricing misreadings
One checkpoint is not the same as one priced prompt. A Build Mode request that changes code normally produces one checkpoint covering the effort, while text-only answers and Plan Mode conversations are also billable. No checkpoint does not establish that no credits were used.
The usage alert is not a cap. It sends an email warning, while the service shutdown limit suspends usage-based services when reached. If predictability is the goal, the second control is the one that enforces it.
The shutdown limit is not Agent-only. Replit credits also cover publishing, storage, databases, and other documented usage, so activity elsewhere can contribute to the same boundary.
Finally, Turbo’s faster advertised response and High effort’s added work do not make either toggle the automatic choice for a difficult-sounding request. Their value depends on the task. Start with the documented mode roles, observe the cost Replit reports, and use that evidence to calibrate the next request without treating a delayed dashboard as the enforcement mechanism.