Cursor "Rate limit exceeded": 4 causes in diagnostic order
Cursor's "Rate limit exceeded" or "Too many requests" error means you've hit one of Cursor's request quotas. Cursor has several different limits — monthly caps, per-minute burst limits, and model-specific quotas — and the fix depends on which one you hit. This page explains each type and what to do about it.
The 30-second answer
- Check your usage first: go to cursor.sh/settings → Usage to see fast requests used vs. monthly cap.
- Monthly cap hit: upgrade your plan or wait for the billing cycle to reset.
- Burst limit hit: slow down — wait a minute and try again.
- Bring your own API key: add your own OpenAI or Anthropic key in Cursor Settings → Models to bypass Cursor's quotas entirely.
Cursor's two types of requests — fast vs. slow
Before diagnosing which limit you hit, understand the distinction:
| Request type | What it is | Monthly cap (Pro) |
|---|---|---|
| Fast requests | Premium models: Claude Sonnet, GPT-4o, etc. Full-speed responses. | 500/month |
| Slow requests | Queue-based delivery when fast quota is exhausted. Same models, slower. | Unlimited (queued) |
| Free tier | Limited fast + slow requests total, no monthly reset | ~50 fast requests total |
When you hit the fast request monthly cap, Cursor doesn't stop — it falls back to slow requests. The error "Rate limit exceeded" usually means you've hit a per-minute burst limit, not the monthly cap.
Cause 1: Monthly fast request cap exhausted
Cursor Pro includes 500 fast requests per billing month. Cursor Tab (inline autocompletion) generates many small requests silently in the background as you type, which can exhaust this quota faster than expected in heavy coding sessions.
How to check:
- Go to cursor.sh/settings.
- Find the Usage section — it shows fast requests used and your cap.
- The month resets on your subscription anniversary, not the 1st of the calendar month.
Fix options:
- Wait for reset: if you're close to the end of your billing cycle, wait it out.
- Upgrade plan: Cursor Business increases the fast request cap significantly.
- Add your own API key: see Cause 3 below — this is the fastest fix.
- Disable Cursor Tab: Settings → Cursor Tab → Disable. You'll still have Cmd+K and Chat, which you control manually and use quota only when you explicitly trigger them.
Cause 2: Per-minute burst limit
Even within your monthly cap, Cursor enforces a per-minute rate limit to prevent burst usage. If you're making many requests rapidly — running an Agent session that loops, repeatedly triggering Cmd+K, or using a script that hammers the Cursor API — you'll hit this limit and get temporary throttling.
Fix: wait 60 seconds and retry. Burst limits are time-windowed and reset automatically. If you're running an agentic session, let it complete one step before triggering the next rather than queuing many actions at once.
Cause 3: Free tier total cap reached
The Cursor free tier has a fixed total number of fast and slow requests with no monthly reset — once used, you're on the free tier indefinitely until you upgrade. The rate limit error on the free tier doesn't mean you can wait and try again; it means the free allowance is gone.
Fix: upgrade to Cursor Pro at cursor.sh/settings, or add your own API key (below) to continue using Cursor without a subscription.
Cause 4 (best long-term fix): Use your own API key
Cursor allows you to bypass its quota system entirely by adding your own model API key. Requests then go directly to the provider (OpenAI, Anthropic, etc.) and are billed at the provider's API rates. This is the most flexible option for heavy users.
How to add your own key:
- Open Cursor Settings (gear icon) → Models.
- Scroll to the API key section for the provider you want (OpenAI, Anthropic, etc.).
- Paste your API key.
- Enable "Use my own API key" — Cursor will use this key for requests to that provider instead of routing through Cursor's quota.
What this costs:
# Approximate API costs (June 2026):
Claude Sonnet 4: ~$3 per 1M input tokens / $15 per 1M output tokens
GPT-4o: ~$2.50 per 1M input tokens / $10 per 1M output tokens
Claude Haiku 4: ~$0.80 per 1M input tokens / $4 per 1M output tokens
For a typical coding session generating ~100K tokens per hour, cost is roughly $0.30–$1.50/hour at current rates — significantly less than upgrading to Business for most individual developers.
FAQ
Does Cursor Tab count against my monthly fast request cap? Yes — each Cursor Tab completion counts as a fast request. On a long coding session, Cursor Tab can consume hundreds of requests per day. If you're hitting the monthly cap, disabling Cursor Tab and using Cmd+K manually is the most effective way to extend your quota.
Can I see which features used my requests? The usage dashboard shows total requests but not a breakdown by feature (Tab vs. Chat vs. Cmd+K). This is a known limitation of Cursor's current usage reporting.
If I add my own API key, does my data go to Cursor or directly to the provider? When you use your own key, Cursor still routes requests through its client but uses your key for authentication. Cursor's privacy policy applies regardless — the model provider also processes your prompts under their privacy terms.
Related
- Cursor authentication failed: 5 causes & fixes
- OpenAI API 429 rate limit vs. quota — how to fix each
- Cursor vs GitHub Copilot: which is better for your workflow?
Last updated June 2026. Cursor plan limits and pricing subject to change — verify current limits at cursor.sh/pricing.