Rate Limits
The per-IP limits on authentication endpoints, how to change one, and how they behave locally and in e2e.
Better Auth limits auth requests per client IP. Counters live in the database (the rate_limit table), so every Worker isolate shares them.
This page is part of the Authentication documentation. Limits for routes that are not sign-in (an upload, a contact form) are in Operations: Rate Limits.
The Limits
| Endpoint | Limit per IP |
|---|---|
/api/auth/sign-in/email | 5 a minute |
/api/auth/sign-up/email | 3 a minute |
/api/auth/change-password | 5 every 10 minutes (the current password is guessed here too) |
/api/auth/link-social | 10 every hour (each one starts a provider redirect) |
/api/auth/request-password-reset | 3 every 5 minutes |
/api/auth/send-verification-email | 3 every 5 minutes |
/api/auth/sign-in/magic-link | 3 every 5 minutes |
/api/auth/magic-link/verify | 5 a minute (the plugin's own) |
/api/auth/email-otp/send-verification-otp | 3 every 5 minutes |
/api/auth/sign-in/email-otp | 5 a minute, and 3 wrong tries per code |
/api/auth/sign-in/social | 10 a minute (each starts an OAuth redirect) |
each /api/auth/two-factor/* endpoint | 3 every 10 seconds (Better Auth's own), 5 wrong codes per sign-in, and a 15-minute lock after 10 in a row (Two-Factor) |
/api/auth/passkey/generate-authenticate-options, /api/auth/passkey/verify-authentication | 10 a minute each (Passkeys) |
/api/auth/organization/create | 5 a minute |
/api/auth/stripe/webhook | 10,000 a minute (Stripe's signed webhooks come from a few addresses, and a burst of billing events must not be refused) |
anything else under /api/auth | 100 a minute |
Per address, whatever the IP: one email address gets at most 10 sign-in codes in 24 hours, counted from its first code (EMAIL_OTP_CODES_PER_ADDRESS, EMAIL_OTP_CODES_WINDOW_HOURS). Each code allows 3 wrong tries, so guesses against one account stay at 30 per window however many IPs an attacker has. It stays on with AUTH_RATE_LIMIT=false, and addresses that aren't valid emails are refused before anything is counted.
Per user, whatever the IP: one user changes their display name at most 20 times an hour (NAME_CHANGES_PER_WINDOW, NAME_CHANGES_WINDOW_HOURS in packages/better-auth/src/constants.ts), over HTTP and from server code (auth.api.updateUser) alike. Email change requests: 3 an hour (EMAIL_CHANGES_PER_WINDOW), an address another account owns included. Saving the same name isn't counted. The Profile page answers the 21st with "You've changed your name too many times. Try again later."
The cost: anyone can use up an address's codes, and its owner then can't get a code until the window ends. With other methods on (password, magic link) they sign in another way; if codes are the only method, that address can't sign in until then. Keep a second method on if that matters to you.
A blocked request gets 429; the forms show "Too many attempts. Wait a minute and try again."
The defaults are AUTH_RATE_LIMIT_RULES in packages/better-auth/src/rate-limit.ts. Plugins added later keep their own limits.
Changing a Limit
Override single rules in your config, inside the object createAuthConfig() returns (replace rateLimitRules: {}); the rest keep their defaults:
apps/web/config/auth.config.server.ts
rateLimitRules: { // many users behind one office IP '/sign-in/email': { window: 60, max: 20 },},window is in seconds. Paths are Better Auth's, without /api/auth, matched exactly.
The Client IP
The IP comes from Cloudflare's cf-connecting-ip header, which Cloudflare sets on every request. x-forwarded-for is ignored: clients can set it.
Per-IP limits slow guessing from one address. An attacker spread over many addresses needs captcha on top: Turnstile on the forms bots abuse.
Turning Them Off
AUTH_RATE_LIMIT=false turns them off. Keep them on when deployed.
Locally every request comes from 127.0.0.1, so a few quick sign-in retries can hit a limit; set AUTH_RATE_LIMIT=false in apps/web/.env if that gets in your way. E2E needs them on: each test sends its own made-up IP (see Testing).