Claude Code Rate Limits Explained
Long Claude Code sessions can stop when an Anthropic usage window is exhausted. The client reports: “You've reached your usage limit. Please wait before sending more messages.”
This guide explains the common limit windows and how routing through an account pool changes that behavior.
How Claude Code rate limits work
Anthropic applies rate limits to Claude Code through multiple overlapping windows. Understanding these is key to knowing why you keep getting locked out.
5-hour rolling window
Every request you make consumes tokens from a 5-hour rolling budget. Once you exhaust your allocation within any 5-hour period, you're locked out until older requests fall outside the window. This is the limit most developers hit during intensive coding sessions.
7-day rolling cap
Even if you pace your 5-hour usage carefully, there's a weekly cap that limits total consumption over a 7-day period. Heavy users typically hit this by Wednesday or Thursday, leaving the rest of the week throttled.
Weekly Opus cap
Claude Opus (the most capable model) has its own separate weekly limit that's more restrictive than Sonnet. If you rely on Opus for complex reasoning tasks, you'll burn through this budget quickly.
Rate limits by plan
Anthropic offers several subscription tiers, each with different limits. Here's how they compare:
| Feature | Free | Pro ($20/mo) | Max ($100/mo) | Max ($200/mo) | OpenLimits (€120/mo) |
|---|---|---|---|---|---|
| 5-hour limit | Very low | Standard | 5x Pro | 20x Pro | Not a user window |
| 7-day cap | Very low | Standard | 5x Pro | 20x Pro | Not a user window |
| Opus access | No | Limited | Higher limit | Highest limit | Every plan |
| Account cooldowns | Hours | Hours | Shorter | Shorter | Managed by pool |
| API access | No | No | No | No | Full REST API |
| Extended thinking | No | Limited | Higher limit | Higher limit | Supported |
| Token analytics | No | No | No | No | Full dashboard |
What happens when you hit a limit
When you reach a rate limit in Claude Code, several things happen:
- Your current request is rejected with a rate limit error
- You're placed in a cooldown period that can last from minutes to hours
- Claude Code may automatically downgrade you to a less capable model (e.g., Haiku instead of Sonnet)
- Extended thinking requests are the first to be throttled
- The interruption breaks context and working momentum
For professional developers, the interruption has a real cost. A 4-hour cooldown in the middle of a feature can disrupt an entire working day.
How OpenLimits handles account limits
OpenLimits is an Anthropic API proxy that pools multiple provider accounts. Instead of routing all your requests through a single account's rate limits, OpenLimits distributes them across a pool of accounts with automatic failover.
How it works
- Your requests are routed through a pool of Anthropic OAuth accounts
- Each account's utilization is tracked across all rate limit buckets (5-hour, 7-day, Opus)
- Users are split evenly across eligible accounts
- If any account hits a rate limit, it's automatically cooled down and traffic shifts to others
- Stable user-account assignment keeps routing predictable while accounts are healthy
This reduces interruptions caused by one account reaching a provider limit. It does not remove the 30 request-start guard or protect against a provider-wide outage.
Setup
Add two environment variables to your shell profile:
export ANTHROPIC_BASE_URL=https://openlimits.app
export ANTHROPIC_API_KEY=your-openlimits-keyEvery configured Claude Code session routes through OpenLimits. Extended thinking, streaming, and tool use continue through the compatible endpoint.
Frequently asked questions
Does OpenLimits work with Claude Code CLI?
Yes. Set the two environment variables above and every Claude Code session will route through OpenLimits automatically. No configuration files to edit.
What models are available?
Every plan includes Claude Opus 4.8 and 4.7, Sonnet 5, and Haiku 4.5. The billing period does not change the Claude catalog.
Is there any added latency?
OpenLimits runs on dedicated VPS infrastructure and preserves streaming behavior expected by Claude-compatible clients.
What about Conductor and other tools?
Any tool that supports ANTHROPIC_BASE_URL works with OpenLimits: Claude Code, Cursor, Conductor, OpenCode, OpenClaw, and any Anthropic SDK client.