Replit’s browser-based AI app builder at replit.com. Agent 3 generates frontend, backend, and database, then deploys to Replit hosting, all from a prompt.
Agent 3 runs autonomous sessions up to 200 minutes (Agent 1 capped at 2 minutes, Agent 2 at 20). It tests its own code, spawns subagents for specialized tasks, and operates across three effort modes: Economy, Power, and Turbo.
Recent developments
- May 1, 2026: Replit argued for independence as Cursor deal talk reshaped AI coding. CEO Amjad Masad told TechCrunch the company is gross-margin positive and wants to remain an end-to-end app platform, not just another AI coding assistant.
System Verdict
Pick Replit Agent if you want an autonomous agent that owns the whole app lifecycle end to end. Agent 3’s 200-minute session length and self-testing loop beats every competitor on sustained unattended work. Everything runs in the browser: editor, terminal, preview, Postgres, deployment. Zero local setup.
Skip it if you need production-grade code quality or architectural control. Generated code is functional, not engineered. Cursor gives a local AI-native IDE with Claude Opus 4.7 for real codebases. Claude Code CLI owns the terminal agent category for working developers.
Who pays which tier: Starter free for exploration with daily AI credits, Core $20/mo annual ($25 monthly) for $25 of credits, 5 collaborators, and up to 2 parallel agents; Pro $95/mo annual ($100 monthly) for $100 of credits, 15 collaborators, up to 10 parallel agents, Turbo mode access, and 28-day database rollbacks. The old Teams tier retired in February 2026. Pro scales credit packs up to $4,000/mo for heavy usage.
Key Facts
| Flagship agent | Agent 3 (2026). Autonomous sessions up to 200 minutes |
| Effort modes | Economy, Power (all plans), Turbo (Pro and Enterprise only) |
| Self-testing | Agent 3 writes, runs, and fixes tests in its own loop |
| Subagents | Agent 3 spawns specialized subagents for isolated tasks |
| Runtime | Replit browser IDE: editor, terminal, preview, database, deploy |
| Database | Replit DB + integrated Postgres |
| Billing | Effort-based credits. Simple changes cost under $0.25 |
| Free tier | Starter with daily AI credits, one published app |
| Top plan | Pro at $95/mo annual ($100 monthly) replaced old Teams tier in February 2026 |
| Parallel agents | Core runs up to 2 agents at once; Pro runs up to 10 |
| Database rollbacks | 28 days on Pro |
| Lite mode | Formerly Fast Mode. Quick scoped changes |
Every data point above was verified against vendor sources on 2026-05-13. See Sources.
What it actually is
A browser-based development environment with an autonomous coding agent layered on top. Agent 3 takes a prompt, writes code, installs packages, provisions a Postgres database, deploys to a public URL, and iterates based on its own test results.
The 200-minute session cap is the headline change versus Agent 2. Agent 3 holds context across longer runs and spawns subagents for isolated work, which is closer to how a real developer breaks down a ticket. Effort modes (Economy, Power, Turbo) let users trade cost for model capability.
The moat is the integrated hosting and database. Everything provisioned by the agent lives on Replit infrastructure with one-click deployment. The weakness is the flip side: apps are bound to Replit, and migrating them out requires rework.
When to pick Replit Agent
- You want an agent that runs unattended for long sessions. Agent 3’s 200-minute cap is the longest in the category.
- You are a non-developer shipping MVPs. The full loop (generate, test, deploy) runs without local tooling or CLI knowledge.
- You teach or learn coding in a classroom setting. Browser-only means no install, no OS-level prerequisites.
- You need a live URL with auth and Postgres in one flow. Hosting, database, and deploy all run on Replit.
- You prefer effort-based pricing over fixed seats. Simple changes cost fractions of a dollar, matching actual work done.
When to pick something else
- Production-grade code on a real codebase: Cursor AI-native IDE with Claude Opus 4.7.
- Terminal agent for working developers: Claude Code CLI, strongest on multi-file refactors.
- Frontend-only in-browser prototyping: Bolt.new with WebContainers.
- Full-stack app with Supabase backend, non-developer friendly: Lovable provisions Postgres and auth automatically.
- Component-level React generation: v0 for shadcn/ui pieces.
Pricing
Pricing via replit.com/pricing. Effort-based credits scale with the work done.
| Plan | Price | Monthly credits | Effort modes | Who’s it for |
|---|---|---|---|---|
| Starter | $0 | Daily AI credits, 1 published app | Economy, Power | Exploration and first prototypes |
| Core | $20/mo annual ($25 monthly) | $25 of credits, 5 collaborators, up to 2 parallel agents | Economy, Power | Most solo builders land here |
| Pro | $95/mo annual ($100 monthly) | $100 of credits, 15 collaborators, 50 viewers, up to 10 parallel agents, 28-day DB rollbacks | Economy, Power, Turbo | Heavy use, teams, power users |
| Pro scaled | $100 to $4,000/mo | $100 to $4,000 credit packs | Full | Agencies and sustained production |
| Enterprise | Custom | Custom plus SSO/SAML, VPC peering, static outbound IPs, single-tenant | Full | Compliance-heavy orgs |
Prices verified 2026-05-13 via replit.com/pricing. Agent usage consumes credits based on effort mode. Credits roll over on Pro. The old Teams tier retired in February 2026, replaced by Pro.
Against the alternatives
| Replit Agent 3 | Lovable | Cursor | |
|---|---|---|---|
| Scope | Full app end to end, hosted | Full app with Supabase backend | Local IDE with AI |
| Autonomy | 200-minute sessions, subagents, self-test | Agent-driven generation with Dev Mode | Cursor Agent, multi-file edits |
| Runtime | Replit hosting | lovable.app hosting | User’s local machine |
| Code quality | Functional, not engineered | Functional, not engineered | Production-grade possible |
| Best for | Non-devs shipping MVPs | Non-devs needing Supabase | Working developers |
| Lock-in | High (Replit infrastructure) | Medium (Supabase + workflow) | Low (local files) |
Failure modes
- Code quality is basic. Generated code works. It is not production-grade. Minimal error handling, sparse tests, opinionated structure. Plan to rewrite before scale.
- Vendor lock-in on hosting. Apps run on Replit. Migrating means untangling Replit DB, Replit auth primitives, and Replit deploy conventions.
- Performance ceiling on Replit hosting. Fine for prototypes and low-traffic MVPs. Sustained production traffic needs proper infrastructure.
- Effort-based cost opacity. “Under $0.25 for simple changes” sets expectations, but complex Agent 3 sessions can burn credits quickly in Turbo.
- Architecture decisions are opinionated. The agent picks stacks, patterns, and data models on its own. Overriding those choices is harder than starting from scratch.
- Browser-only runtime constrains workflows. No local tooling, no native debugger, no OS-level packages beyond what Replit exposes.
- Turbo mode is Pro-gated. Starter and Core cannot access the fastest models. Expect slower Agent 3 runs on lower tiers.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility × Value × Moat × Longevity, unweighted average). Last verified 2026-05-13 against replit.com/pricing, the Replit Pro launch post, the effort-based pricing post, docs.replit.com, and the May 1, 2026 Cursor-deal independence coverage.
FAQ
Is Replit Agent free? Yes. The Starter tier gives daily AI credits, one published app, and access to Economy and Power effort modes. Enough to test the agent, not enough for sustained use (replit.com/pricing).
What changed with Agent 3? Agent 3 runs autonomous sessions up to 200 minutes (versus 2 minutes on Agent 1 and 20 minutes on Agent 2). It tests and fixes its own code, spawns subagents for specialized tasks, and operates across three effort modes: Economy, Power, Turbo (Replit Pro launch).
How do effort modes work? Economy runs slowest and cheapest, Turbo runs fastest and most expensive. Economy and Power are on all plans. Turbo is Pro and Enterprise only. The same prompt on Turbo costs more credits but completes faster with stronger model routing.
Replit Agent vs Cursor? Replit Agent builds and deploys full apps end to end on Replit hosting, targeted at non-developers. Cursor is a local AI-native IDE for working developers who want architectural control. Pick Replit if you want apps built. Pick Cursor if you want to write and control code.
What happened to the Teams plan? Retired in February 2026. Pro at $95/mo annual ($100 monthly) replaced it with 15 collaborators, 50 viewers, Turbo access, up to 10 parallel agents, 28-day database rollbacks, and credit rollover.
Sources
- replit.com/pricing: current tier list and credit allocations
- blog.replit.com/pro-plan: Pro launch and Teams retirement
- blog.replit.com/effort-based-pricing: effort-based credit model
- docs.replit.com/billing/ai-billing: effort mode mechanics
Related
- Category: AI Coding
- Guides: Best Cursor Alternatives (2026)
