AI Coding Tool Rate Limits Explained: How to Avoid Getting Blocked Mid-Sprint

You’re sprinting toward a deadline, the AI Agent is mid-task, and suddenly: “rate limit exceeded.” Here’s how each major AI coding tool’s limits work — and how to stop them from derailing your flow.

---

Rate Limits by Tool

Claude Code

Plan	Rate Limit
Pro ($20/mo)	~45 requests / 5-hour rolling window
Max 5x ($100/mo)	~225 requests / 5-hour window
Max 20x ($200/mo)	Higher limit + full Opus 4.6 access
API (pay-as-you-go)	Anthropic API tiers (Tier 1–5)

Important: The “5-hour window” is a rolling window, not reset at the top of the hour. Starting from your first request, you have a 5-hour period during which the total can’t exceed the limit.

Cursor

Cursor Pro uses a credit pool, not a request count:

• Pro: $20 credit pool/month
• Pro+: $70 credit pool/month
• When the pool is empty, usage continues at model cost rates ($0.04–$0.50/request) — you won’t get a hard stop, you’ll just get charged more
• No “rate limit exceeded” error in Auto mode — empty pool = extra billing, not a block

Cline (with Anthropic API)

Cline itself has no rate limits. Limits come from your chosen API provider. With Anthropic API directly:

• New accounts (Tier 1): 60 RPM, 100K TPM — fairly low
• Limits auto-upgrade after adding funds to your account
• With Volcengine Ark or local Ollama: essentially no rate limits

Windsurf

• Credit-based (500 credits/month on Pro)
• Additional credits at $10/250 credits (add-ons never expire)
• No request frequency hard limit; exhausting credits stops access to premium models

GitHub Copilot

• Pro: 300 premium requests/month, $0.04/request overage
• Pro+: 1500 premium requests/month
• Code completion is unlimited and not rate-limited

---

Why Rate Limits Always Hit at the Worst Moment

Common patterns:

1. Continuous large task: Agent handling a major refactor auto-generates 30+ tool calls, consuming the 5-hour window before you realize it
2. Forgotten /compact: Claude Code conversation history grows, each request carries more context, effectively accelerating window consumption
3. Month-end crunch: Cursor credit pool exhausted, Copilot monthly quota used up
4. Multi-project use: Using the same account across multiple projects — requests from all projects count against the same limit

---

5 Strategies to Handle Rate Limits

Strategy 1: Keep a Backup Tool Ready

The most effective approach: when your primary tool hits a limit, switch to a backup immediately without breaking your flow.

Recommended pairings:

• Primary: Claude Code Pro → Backup: Cline + Anthropic API (same model, pay-as-you-go)
• Primary: Claude Code Pro → Backup: Cline + Volcengine Ark (official promo pricing, China-friendly)
• Primary: Cursor Pro → Backup: Windsurf Free (25 credits/mo) or Cline + Ark
• Primary: Copilot Pro → Backup: Cursor Hobby or Cline

Cline configurations are saved — switching takes seconds.

Strategy 2: Pace Your Requests

Claude Code Pro’s 45 requests/5-hour window works out to one request every ~6.7 minutes on average.

Adapt your workflow to this rhythm:

• While Agent is executing, do code review or documentation work
• Don’t send rapid-fire follow-up messages — describe the problem clearly in one well-crafted request
• For complex tasks, use Plan mode to confirm the approach first, then execute — this reduces back-and-forth

Strategy 3: Try Pay-as-You-Go Before Upgrading

Many Claude Code users immediately consider upgrading to Max 5x ($100/mo) when they hit limits. But:

• If you only hit limits on 2–3 days per month, paying $0.15–0.50 per extra request beats paying $80 more every month
• Check if Claude Code’s settings let you enable “continue on API billing after limit” — this auto-switches to pay-as-you-go seamlessly when the window is full

Strategy 4: Use /compact to Slow Context Growth in Claude Code

Longer conversation history = more tokens per request = faster window consumption.

# Run in the Claude Code conversation:
/compact

Run /compact after completing each stage of a task. It compresses history into a summary, letting you handle more tasks within the same 5-hour window.

Strategy 5: Batch Non-Urgent Tasks to Off-Peak Hours

Some AI coding tasks don’t require real-time interaction:

• Bulk code reviews
• Writing documentation / comments
• Generating test cases
• Code formatting and minor refactors

Schedule these for evenings or early mornings. This preserves your rate window capacity for urgent daytime work and lets you handle batch tasks more comfortably.

---

What to Do When You Hit a Limit Right Now

1. Switch to a backup tool immediately — Cline + Ark or API, don’t just wait
2. Note when it happened — patterns usually emerge (same time of day, same type of task)
3. Assess whether you need to upgrade: consistently hitting limits during work hours → current plan is too small
4. Claude Code users: Check settings for “continue on API billing” mode to avoid hard stops

---

Rate Limit Summary

Tool	Limit Type	When Limit Hits	Recovery Cost
Claude Code Pro	45 req/5-hour	Wait for window reset or switch	Upgrade +$80/mo
Claude Code Max 5x	225 req/5-hour	Wait or use API mode	Add API credits
Cursor Pro	$20 pool/month	Auto-charged at model rates	Per-request billing
Cline + Anthropic API	API Tier	Wait or add funds	$10–50 top-up
Cline + Volcengine Ark	Essentially unlimited	—	—
Windsurf Pro	500 credits/month	Add-on credits $10/250	Flexible top-up
Copilot Pro	300 premium/month	$0.04/extra request	Low overage cost

---

Related

• How to Save Money on Cursor
• Claude Code Token Saving Guide
• Claude Code Is Too Expensive: Budget Alternatives
• Cline + Volcengine Ark Setup Guide