Replit AI pricing

About

Replit is a browser-based coding workspace built around Replit Agent — an AI that builds and deploys full applications from a prompt. You describe what you want, the Agent writes the code, runs it, and ships it to a live URL, all inside the same workspace. That “prompt to deployed app” loop is what Replit now sells, and it reshaped how the company prices.

Founded in 2016 and headquartered in San Francisco, Replit started as an online IDE and REPL for learners and hobbyists. The Agent pivot turned it into one of the most visible “vibe coding” platforms, used by non-engineers and developers alike to spin up business apps, prototypes, and internal tools. The company has raised from investors including a16z and Coatue and reported a sharp jump in revenue through 2025 as Agent usage scaled.

The pricing question Replit has to answer is unusual: an AI agent’s cost is genuinely variable — a one-line fix and a multi-file refactor are not the same compute — so a flat seat price alone can’t cover it. Replit’s answer is a hybrid: you buy a seat, the seat comes with a prepaid dollar wallet of credits, and the Agent (plus your deployments) draws that wallet down by effort.

For the current rate card, see Replit’s pricing page.

Pricing summary : How Replit’s pricing model works

Replit’s pricing has three layers stacked on top of each other:

1. A seat tier — Free Starter, Core at $20/mo (annual; $25 monthly), Pro at $95/mo (annual; $100 monthly), or custom Enterprise. The seat sets your collaborator count, parallel-agent limit, compute size, and model access.
2. An included credit wallet — every paid tier bundles a fixed dollar allotment of credits: $25 of monthly credits on Core, $100 on Pro. These are spent on Agent/Assistant work and deployment compute.
3. Effort-based usage — once the wallet is empty, you pay pay-as-you-go. Replit Agent and Assistant are billed by effort: each request is one checkpoint priced on the time and compute it actually used.

So the headline price is not your real bill. Core’s $20 buys a seat plus a $25 credit wallet; whether you ever pay more depends entirely on how hard you lean on the Agent and how much always-on compute your deployed apps consume.

What makes this different: Replit bundles a dollar-denominated credit allowance into a seat price and meters AI work by genuine effort rather than a flat per-task fee. It’s closer to a phone plan with included minutes than to a classic per-seat SaaS subscription.

Pricing by product

Across all paid tiers, the Agent and Assistant draw down the included credit wallet first, then bill pay-as-you-go by effort. The compute size attached to each seat also rises with the tier (Starter is smallest; Pro gives the largest workspace).

Sales motions across products: Starter, Core, and Pro are fully self-serve / PLG (sign up and pay by card). Enterprise is sales-led — custom seat limits, SSO/SAML, single-tenant environments, and VPC peering are quoted by Replit’s sales team.

Hidden costs : What Replit users actually pay

The seat price is the floor, not the bill. Three things push real spend above it, and none are visible on the plan card:

• Agent effort beyond credits. Each request is one checkpoint, priced on time and compute. Simple changes typically cost less than $0.25; a full feature build runs about $1 to $3 in Agent effort (second-source: Replit billing docs and 2026 pricing analyses). A heavy day of building blows through Core’s $25 wallet fast.
• Always-on deployments. Static deploys are cheap, but an always-on reserved VM typically runs $5 to $20+/month per app, and autoscale deployments bill by use. These are charged against credits / pay-as-you-go on top of your seat.
• Database and storage. Replit’s managed PostgreSQL runs about $1.50 per GB-month plus $0.16 per compute-hour; app storage is roughly $0.03 per GB-month and outbound data transfer about $0.10 per GB.

Put together, heavy Agent users routinely report $100 to $300 per month on top of their base plan.

Want to estimate your own Replit bill? Use the Replit AI pricing calculator to model seat cost plus Agent effort and deployment compute.

For the mechanics behind this, see our guide on usage-based pricing fundamentals and our analysis of bill shock and cost unpredictability.

Pricing evolution : Replit pricing history and changes

Cadence

Tracked range: 2025–present. History reconstructed from Replit’s effort-based-pricing and Pro-plan announcements; Wayback access was unavailable from this environment.

Notable changes

• 2025-06-18 — Replit announced effort-based pricing for Replit Agent (rollout from July 1). The previous flat $0.25-per-checkpoint rate was replaced by per-request pricing that scales with the time and compute each task consumes, alongside new High Power model and Extended Thinking controls.
• 2026-02-24 — Replit launched the Pro plan as its serious-builder tier, dropped Core from $25 to $20/mo (annual), and began sunsetting the Teams plan, auto-upgrading Teams subscribers to Pro starting March 3, 2026.

This is a notable direction: a price cut on the mainstream tier (Core) paired with a new higher tier (Pro) and a usage meter (effort-based) that captures the variable AI cost. See how AI companies are shifting away from per-user licenses for the broader pattern.

What’s unique : Replit’s distinctive pricing mechanics

1. Dollar-denominated credit wallets inside a seat price. Most SaaS bundles a usage quota (X requests, Y seats). Replit bundles a literal dollar allowance — $25 on Core, $100 on Pro — that the Agent and deployments spend down. The seat price is part subscription, part prepaid wallet.

2. Effort-based Agent metering. Replit abandoned a flat per-checkpoint fee because an AI agent’s real cost is variable. Now each request is one checkpoint priced on the time and compute it consumed, so a trivial edit is cheap and a complex multi-file task costs more — with user-facing High Power and Extended Thinking dials to control it.

3. A price cut as a strategic move. Dropping Core from $25 to $20 while adding a $95 Pro tier is unusual. Replit widened the funnel at the bottom (cheaper Core) and captured serious builders at the top (Pro), letting the effort meter — not the seat — absorb heavy AI usage.

Strengths & weaknesses

Billing UX : Replit billing controls and transparency

• Billing controls — Replit exposes per-request effort controls (High Power model, Extended Thinking) so you can dial Agent spend up or down on each task, plus account-level spend visibility. Pricing is fully public: Starter, Core ($20/mo annual), and Pro ($95/mo annual) all show their rates and included credits on the pricing page.
• Usage visibility — Credit consumption is tracked in-product, and the billing docs explain how Agent/Assistant effort and deployment compute draw down credits. Pro adds credit rollover (one month) and a tiered credit-purchase ladder for teams that need more headroom.
• Payment options — Self-serve plans are card-billed monthly or annually (annual saves up to 20% on Core). Enterprise supports invoicing and custom terms. Replit notes prices are subject to local tax.

For how to design controls like these, see our guide on usage invoicing and billing cycles.

Strategic wins : Why Replit’s pricing decisions worked

1. Bundling a credit wallet to absorb variable AI cost

By putting a $25 (Core) / $100 (Pro) dollar wallet inside the seat, Replit made the seat price feel complete for light users while still metering the genuinely variable Agent cost underneath. Most casual builders never exceed the wallet, so the headline price holds — but power users self-fund their own compute. See choosing the right usage metric for why a dollar wallet beats a raw task quota here.

2. Switching from flat checkpoints to effort

The old $0.25-per-checkpoint model under-charged complex tasks and over-charged trivial ones as the Agent got more capable. Effort-based pricing realigned revenue with cost and gave users dials to manage it — a textbook example of how AI companies structure pricing around real compute. Related: outcome-based pricing trends.

3. Cutting Core while adding Pro

Lowering Core to $20 widened the top of the funnel exactly as the “vibe coding” market exploded, while Pro captured the serious builders Teams used to serve. The effort meter — not the seat — carries the heavy-usage revenue, so Replit could afford to make the entry seat cheaper.

Areas to improve : Gaps in Replit’s pricing approach

1. Spend predictability

Effort-based billing is honest but opaque: “how many features is $25 of credits?” has no fixed answer, and community threads regularly cite surprise bills. Clearer per-action estimates before a checkpoint runs would reduce bill shock.

2. Credit mechanics clarity

Credits are dollar-denominated and Core’s don’t roll over, while Pro’s roll one month — a subtle distinction that’s easy to miss at signup. Surfacing rollover and burn-rate prominently would help buyers compare tiers honestly.

3. Stacked compute costs

Deployment, database, and storage compute all bill on top of the seat and credits. Bundling a baseline deployment allowance into Core, or showing a live “projected monthly total,” would make the true cost of shipping an app less of a surprise.

Key takeaways

1. Replit’s real price is seat + effort + compute. Core is $20/mo (annual) with a $25 credit wallet, but Agent effort and deployments determine the actual bill.
2. Credits are a dollar wallet, not a quota. Core bundles $25, Pro bundles $100 — spent on Agent work and compute, with Pro rolling over one month.
3. Effort-based metering replaced flat checkpoints. Since June 2025, each request costs what it actually consumes, with High Power and Extended Thinking dials.
4. Core got cheaper in 2026. A $25 → $20 cut, paired with the new $95 Pro tier and the Teams sunset, widened the funnel while the meter caught heavy use.
5. Predictability is the trade-off. Honest effort pricing means variable bills; heavy users report $100–$300/mo on top of base.

UBP implications

1. Bundle a dollar wallet to make usage pricing feel like a subscription. Replit’s included credits let light users see a flat price while still metering variable AI cost — a clean pattern for any usage-based pricing launch worried about sticker shock.
2. Meter by effort when cost is genuinely variable. A flat per-task fee mis-prices an AI agent; effort-based checkpoints align revenue with compute. Pick a usage metric that tracks your real cost driver.
3. You can cut the entry seat if a meter carries the heavy use. Replit lowered Core because effort billing — not the seat — captures power-user revenue, a reminder that seat price and usage meter are separate levers.

Sources

• Replit pricing page (accessed 2026-06-16)
• Replit blog — Introducing Effort-Based Pricing for Replit Agent (accessed 2026-06-16)
• Replit blog — Replit Pro Is Here, and Core Now Offers Even Better Value (accessed 2026-06-16)
• No Code MBA — Replit Pricing 2026 (accessed 2026-06-16)
• We Are Founders — Replit Pricing: What You Actually Pay (accessed 2026-06-16)

Browse more fully-researched profiles in the pricing blueprint.

Bottom line

Replit’s $20 Core seat is a doorway, not a ceiling. The real model is a seat bundled with a dollar credit wallet, an effort-metered Agent that bills by the compute each request consumes, and pay-as-you-go deployment costs on top — which is why light builders pay the sticker price and heavy ones pay multiples of it.

Want to compare Replit against other AI coding tools? Browse the pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Pro plan launches; Core drops to $20, Teams sunset

Feb 2026

Replit launched the Pro plan ($95/mo annual, $100 monthly credits) as its serious-builder tier, dropped Core from $25 to $20/mo (annual), and began sunsetting the Teams plan, auto-upgrading Teams users to Pro.

Effort-based pricing replaces flat $0.25 checkpoints

Jun 2025

Replit announced effort-based pricing for Replit Agent on June 18, 2025 (rolled out July 1). The old flat $0.25-per-checkpoint rate was replaced by per-request pricing that reflects the time and compute each task consumes, plus High Power and Extended Thinking controls.