Replit pricing explained: plans, credits and hosting costs
Replit pricing has two parts: a plan that includes monthly credits, and usage billed against those credits. Starter is free, Core is $20 a month with $20 of credits, Pro starts at $100 a month with at least $100 of credits, and Enterprise is priced on contract. The same credits pay for building with Agent and for running your published app, so once a busy month uses them up, the rest of your hosting and database usage is billed pay-as-you-go.
The figures below come from Replit’s pricing page and billing docs. Replit changes plans often, so check both before you budget.
Replit plans and what they include
| Plan | Price | Monthly credits | What else |
|---|---|---|---|
| Starter | Free | Daily Agent credits up to a monthly cap, plus monthly cloud credits | Lite build only. One published app, which expires after 30 days unless you republish it. Production database limited to 1 GiB |
| Core | $20 a month, or $18 billed yearly | $20 | Full Agent, a Free Mode allowance, unlimited published apps, 5 collaborators, 100 GiB of outbound data for deployments |
| Pro | From $100 a month, or $90 billed yearly | At least $100, in tiers. Unused credits roll over for two months | 15 builders sharing one credit pool, 10 parallel Agent tasks, 28-day database restore window |
| Enterprise | Custom | Custom | SSO, SCIM, audit logs, invoicing |
Pro replaced the old Teams plan and has no per-seat fee. Core credits reset on your renewal date and don’t roll over. If you’re still choosing a builder, our Replit vs Lovable comparison covers how the two are priced.
How Replit Agent usage is billed
Agent has three modes. Free Mode costs nothing within your plan’s allowance, which refreshes every five hours on Core and Pro. Power and Max use credits, and Replit asks you to confirm before a request moves from Free Mode into a paid mode.
Paid work uses what Replit calls effort-based pricing. Each request ends in one checkpoint (a saved snapshot of your app), and the price of that checkpoint grows with how long the Agent worked. A small bug fix costs less than a whole feature. Replit doesn’t publish a fixed price per checkpoint; you see each one’s cost by hovering over the usage icon in the Agent tab. Two details catch people out:
- Plan Mode can bill without changing code. In the paid modes, planning is reasoning, and reasoning costs credits.
- Third-party AI is billed at the provider’s price. When Agent uses models such as Claude, ChatGPT or Gemini behind the scenes, the provider’s public API rate comes out of your credits.
A checkpoint is billed when the Agent finishes the request, and rolling back afterwards doesn’t refund it. The cheapest Agent request is one scoped well enough that you keep the result.
How Replit hosting is billed
Publishing an app is called a deployment, and you choose one of four types.
| Type | What you pay | Good for |
|---|---|---|
| Autoscale | $2 a month base fee, $0.60 per million compute units, $0.40 per million requests | Web apps and APIs with uneven traffic. Scales to zero when idle |
| Reserved VM | Fixed monthly: $15 for a shared 0.5 vCPU / 2 GB machine, $35 for dedicated 1 vCPU / 4 GB, $50 for 2 vCPU / 8 GB, $130 for 4 vCPU / 16 GB | Always-on servers, bots, background workers |
| Scheduled | $2 a month base fee, $0.60 per million compute units while each run lasts | Cron-style jobs such as reports and cleanups |
| Static | Free hosting, $0.05 per GB served | Landing pages and docs. Not for Agent-built full-stack apps, which need a server |
Compute units measure CPU and memory over time: one CPU-second is 18 units and one GB-second of memory is 2. Autoscale only charges while requests are being served, and the machine size you pick decides how many units each second costs.
Around the deployment, these lines draw on the same credits:
| Usage | Price |
|---|---|
| Outbound data (data your app sends out) | 100 GiB included on Core, then $0.05 per GiB. Incoming data is free |
| Production database compute | $0.16 per hour the database is active |
| Production database storage | $0.35 per GiB per month, billed on the month’s peak, including history |
| App Storage (files) | $0.015 per GiB per month, $0.05 per GiB transferred, plus small per-operation fees |
| Replit AI Integrations (AI calls your app makes) | The provider’s public API price |
The development database you build against is free. Replit cut most hosting prices in August 2026, so older guides quoting $3.20 per million compute units are out of date.
What a small app costs on Replit
Here is one month for a small web app on Core, using the published prices above. Assume an Autoscale machine with 1 vCPU and 2 GB of memory, busy serving requests for 6 hours a day, 200,000 requests, a production database active 12 hours a day, 1 GiB of data, 10 GiB of uploaded files and 40 GiB of outbound data.
| Line | Usage | Cost |
|---|---|---|
| Autoscale base fee | 1 deployment | $2.00 |
| Compute units | 180 busy hours × 22 units a second = 14.3 million | $8.55 |
| Requests | 200,000 | $0.08 |
| Database compute | 360 active hours | $57.60 |
| Database storage | 1 GiB | $0.35 |
| App Storage | 10 GiB | $0.15 |
| Outbound data | 40 GiB, within the 100 GiB included | $0.00 |
| Total usage | about $69 |
Core’s $20 of credits covers the first $20 of that, so you pay the $20 plan plus about $49 of usage: about $69 in total, before any Agent work. The server is the small line. The database is most of the bill, because it stays active for five minutes after every query, so steady traffic keeps it awake far longer than the server is busy.
Two benchmarks from the same prices. A database kept awake all month costs about $117 in compute alone (730 hours × $0.16). An Autoscale machine of that size that’s busy around the clock costs about $35 a month, the price of a dedicated 1 vCPU Reserved VM with twice the memory, so an app that’s busy all day gets more for its money on a Reserved VM.
What makes a growing Replit bill jump
- A database that never sleeps. Anything that queries every few minutes, such as an uptime check, a dashboard that polls, or a scheduled job, keeps the database active and billing every hour of the month.
- Autoscale that never scales to zero. Bots, crawlers and front-end polling keep requests flowing. Each extra machine Autoscale adds bills in parallel, up to the Max machines limit you set.
- Agent and hosting sharing one pool. A heavy build month can spend the credits that would have covered hosting, so the hosting goes to pay-as-you-go.
- AI features in your app. When you ask Agent for AI functionality, it defaults to Replit AI Integrations, so every model call your users trigger is charged to the same credits. With your own API key, the provider bills you instead.
- Media. Downloads past the 100 GiB allowance cost $0.05 per GiB, and files served from App Storage are billed for transfer as well.
How to control your Replit spending
- Open the Usage page and filter by resource and project to find your biggest line. Data can take up to 30 minutes to appear.
- Compare PostgreSQL Compute hours with the hours in the month. If they’re close, something is keeping the database awake.
- Check your deployment settings. Set Max machines on Autoscale, and price a Reserved VM if your app is busy most of the day.
- Set usage alerts before a budget. They email you when spending beyond your credits passes an amount you choose.
- Set a budget deliberately. A budget is a hard cap: when it’s reached, Replit suspends usage-based services until the next cycle or until you raise it, and that includes your published app. Set it high enough that a good month doesn’t take you offline.
- Use Free Mode for everyday Agent work, and switch to Power or Max only for work that needs them. Planning in a paid mode costs credits too.
If you need more credits, Core and Pro accounts can buy packs: $100, $300 for $290, $500 for $480, or $1,000 for $950. Packs expire six months after purchase, and auto-reload can buy a new one when your balance runs low.
When to move production hosting off Replit
We moved an AI design platform for businesses off a Replit Autoscale deployment and onto two Render services: staging auto-deploys the staging branch, production auto-deploys main, and features reach staging by pull request. That meant replacing Replit Auth with Supabase Auth and moving files from Replit’s object storage to Cloudflare R2. The founder kept prototyping in Replit on his own branch, and we merged that work into staging before release.
Moving makes sense when you have a concrete reason:
- You need a reviewed release process with separate staging and production environments and engineers working in their own tools.
- Your usage is always on. If the database and server never sleep, compare a fixed-price setup, starting with Replit’s own Reserved VM.
- You depend on Replit-only services such as Replit Auth and want sign-in and data in accounts your team controls.
Leaving isn’t the default answer. Replit keeps production and development databases separate and offers point-in-time database restore, and plenty of apps run well on Replit once the code is solid. Our guide to taking a Replit app to production covers what to fix first, and Replit alternatives compares where to go if you leave.
FAQ
Is Replit free?
The Starter plan is free, with daily Agent credits up to a monthly cap and one published app that expires after 30 days unless you republish it. Publishing more than one app, or keeping one online without republishing, needs Core or above.
Do unused Replit credits roll over?
Not on Core: they reset on your renewal date. Pro credits roll over for two months. Credit packs expire six months after purchase.
Why is my Replit bill higher than my plan price?
Your plan price buys credits, and Agent work, hosting, the production database, storage, data transfer and AI calls all draw on them. Anything beyond the credits is billed pay-as-you-go up to your budget. The Usage page shows which line went over.
If your Replit app is ready for paying customers and you’re not sure what it should cost to run or where it should live, our Replit rescue starts by mapping exactly that.
