What Are Claude's Rate Limits in 2026? (Complete Guide)

If you've ever been mid-task in Claude and hit a rate limit without warning, you know how disruptive it is. Claude's limits vary significantly depending on whether you're using claude.ai (the chat interface), the Claude API directly, or Claude Code. Here's a complete breakdown for 2026, including exact numbers, how the rolling window works, and what you can do to stay ahead of the limits.

claude.ai Rate Limits by Plan

The claude.ai web and mobile interface uses a message-based limit inside a rolling 5-hour window. "Rolling" means the window doesn't reset at a fixed time like midnight — it resets 5 hours after your first message in the current period. This is the detail most people miss, and it's why getting rate-limited can feel inconsistent.

Claude Free

The free tier gives you access to Claude Haiku with a moderate daily message allowance, plus limited access to Sonnet. Exact numbers aren't published by Anthropic, but active users typically hit throttling after 20–30 Sonnet messages per day. Haiku messages are more generous. For light personal use, the free tier is sufficient; for development work, it's usually not.

Claude Pro ($20/month)

Pro users get roughly 45 messages per 5-hour rolling window on Claude Sonnet 3.5 and 3.7. Haiku messages don't count against this limit. The 45-message ceiling is an approximation — Anthropic adjusts it based on overall platform load, so you may occasionally see slightly higher or lower limits. The rolling window mechanic means your reset time is unpredictable if you don't track it: it depends on when you started your current session, not the clock.

Claude Max 5× ($100/month)

Approximately 225 Sonnet messages per 5-hour window. Same rolling window mechanic as Pro. Designed for heavy users who regularly hit the Pro ceiling during long coding or writing sessions.

Claude Max 20× ($200/month)

Approximately 900 Sonnet messages per 5-hour window — which for most practical purposes means you won't hit it during normal usage. At this tier, the limit stops being a daily concern and only comes up during extremely intensive automated workflows.

Claude API Rate Limits

The Anthropic API uses a completely different limiting system from claude.ai. Instead of message counts in 5-hour windows, API limits are expressed as:

TPM (tokens per minute): how many tokens you can process per minute across all requests
RPM (requests per minute): how many API calls you can make per minute

These limits vary by model and by your organization's API tier (Tier 1 through 4). New organizations start at Tier 1 with conservative limits. Tiers increase automatically as your cumulative API spend reaches thresholds — roughly $100 for Tier 2, $500 for Tier 3, and so on. Higher tiers unlock higher TPM and RPM limits.

Model-specific limits also apply. Claude Haiku has the highest RPM limits and the lowest cost. Claude Sonnet has mid-tier limits and is the most common choice for production applications. Claude Opus has the lowest RPM limits and highest cost per token.

Claude Code Rate Limits

Claude Code is a CLI tool that calls the Anthropic API directly. It is not limited by the claude.ai 5-hour message window — instead, it draws from your organization's API token budget, billed monthly.

The key difference that trips up many developers: a single Claude Code agentic loop that reads files, runs shell commands, evaluates results, and iterates on edits can consume tens of thousands of tokens in a few minutes. A 30-minute debugging session with Claude Code can cost more in tokens than an hour of claude.ai Pro chat, simply because of how much context gets passed back and forth.

Claude Code defaults to Claude Sonnet for most operations, which is mid-cost. Using Claude Haiku where possible reduces API spend significantly. See our guide on how to avoid Claude Code rate limits for practical strategies.

The Critical Difference: Chat vs Code Limits Are Separate

This is the most important thing to understand. Your claude.ai 5-hour message window and your Claude Code API token budget are completely independent systems. Hitting one doesn't affect the other. Upgrading to Claude Max raises your chat limits but does nothing for your Claude Code capacity — that depends on your Anthropic API tier, which is managed separately.

Developers who use both products in the same workflow are effectively managing two separate quota systems simultaneously, with no native unified view.

For a detailed comparison, see Claude Code vs Claude Web: Rate Limit Differences.

Why Claude's Native Indicators Aren't Enough

The claude.ai interface shows a vague "you're approaching your limit" indicator, but it doesn't tell you: how many messages you have left, how many minutes until the window resets, or anything about your Claude Code API usage. There's no push notification, no proactive warning that lets you wrap up cleanly before the limit hits.

For the API, the Anthropic Console shows usage dashboards, but they update with a delay and aren't designed for ambient monitoring while you're in the middle of a coding session.

How to Track Claude Rate Limits Automatically

AIUsageBar is a native macOS menu bar app that shows your live Claude usage data at a glance: current session message count, the percentage of your 5-hour window used, exact minutes until reset, and API token spend for Claude Code. It updates in real time so the information is always current when you glance at your menu bar.

The free tier covers Claude-only tracking. If you also use OpenAI, Cursor, Copilot, or Gemini, the Pro plan ($9.99 one-time) adds all five providers to the same popup — so you can see at a glance which tool has capacity remaining before you commit to a long task.