Search topic
claude code rate limits
Practical outcome
The weekly and 5-hour caps demystified — plus how to see your real utilization.
Quick answer
You've hit your session limit or You've hit your weekly limit with the reset time. That's not an HTTP 429: a 429 means an API-key or cloud rate limit. Run /usage for both bars, and npx whoburnedmore --local for the daily token history behind them. 📊Most developers assume Claude Code has one rate limit, with one number and one reset. There are really two independent clocks on a subscription, and a third, different system on the API. You can be well inside your 5-hour window and still be blocked because the weekly total ran out. Anthropic's docs warn that one burst, such as a large workflow fan-out, can use up the weekly allowance before the session window resets. This guide explains how each limit works as of November 8, 2026, what the messages mean, and how to estimate your own headroom.
How do the two rate-limit windows work?
The 5-hour session limit starts when you send your first message. According to Anthropic's help center, the limit “resets in full five hours later.” It is a fixed block from that first message, not a sliding window. Usage from claude.ai, the desktop app, IDE extensions and Claude Code all counts against it. Pro and Max have it. Verified Ultra accounts don't. Unverified Ultra accounts get a temporary one until automatic verification finishes. If Settings > Usage shows no “Current session” bar, you're verified.
The weekly limit is a separate, larger cap across all models. It resets at a fixed time each week that is assigned to your account. That time doesn't depend on when you subscribed or started working. Max and Ultra also have a separate weekly Fable limit, with Fable included up to half the weekly limit. Anthropic announced weekly limits on July 28, 2025, and they took effect on August 28, 2025. After a May 13 to September 13, 2026 promotion that raised them by 50%, weekly Claude Code limits have been 25% higher than before the promotion since September 14, 2026.
How big is each plan's allowance?
Anthropic publishes multiples, not token counts. Max gives 5× Pro's usage per 5 hours and per week. Ultra, formerly Max 20x, gives 10× Pro per week with no 5-hour limit on verified accounts. Team Premium seats get 5× Standard seats. How fast you use it depends on the model, the length of your conversations, file sizes and the effort level. Opus at high effort on a long context burns through a window far faster than Sonnet on a fresh one. Some accounts also see model-specific messages such as You've hit your Opus limit. Those block only that model family, so /model to a different family keeps you working.
Clearing the 5-hour window does not clear the weekly cap
These limits stack. If you hit the weekly cap, waiting five hours won't help, and neither will switching models: both the session and weekly limits apply across all models. Each message prints its own reset time, which tells you which clock you hit:resets 3:45pm for the session, resets Mon 12:00am for the week. 🔥What triggers a rate-limit block and what does it look like?
Every request Claude Code sends draws on your allowance. That includes your prompts and each follow-up request after a tool call, plus subagents, workflows, /loop tasks and compaction. Before a window runs out, Claude Code warns you. At the limit it stops and names the window. These are the messages from Anthropic's error reference:
You've used 85% of your session limit · resets 3:45pm You've hit your session limit · resets 3:45pmYou've hit your weekly limit · resets Mon 12:00amYou've hit your Opus limit · resets 3:45pm # not your plan quota:API Error: Server is temporarily limiting requests (not your usage limit)API Error: Request rejected (429) · this may be a temporary capacity issue. …
The last two lines are different problems. “Server is temporarily limiting requests” is a short server-side throttle, and Claude Code retries it automatically. Request rejected (429) means you hit the rate limit on an API key, an Amazon Bedrock project or a Google Cloud project. A common cause is a stray ANTHROPIC_API_KEY in your shell. Claude Code then bills that key at API rates instead of using your subscription, so check the active credential with /status. A 529 Overloaded error is a capacity problem and doesn't count against your quota. Claude Code retries 429 and 529 errors with exponential backoff, up to 10 times by default (CLAUDE_CODE_MAX_RETRIES).
How do I calculate my real utilization?
/usage shows each window as a percentage with its reset time. A custom status line can show the same numbers continuously, from rate_limits.five_hour.used_percentage, rate_limits.seven_day.used_percentage and the matching resets_at fields. None of these give token counts. To turn a percentage into something you can plan with, compare it with what you actually burned in the same window:
For token counts per 5-hour block, npx ccusage blocks groups Claude Code usage into 5-hour windows locally. For the longer picture, npx whoburnedmore --local shows daily totals by model. Treat the result as your own yardstick, not Anthropic's formula. Plan usage isn't a raw token count. It depends on the model, and Anthropic says re-read history draws on your plan at the cached-token rate, not the full input rate.
API keys: a different system entirely
If Claude Code runs on a Console API key, there are no 5-hour or weekly windows. Your organization has per-minute limits by usage tier, enforced with a token bucket that refills continuously, and a monthly spend cap. Cached input doesn't count toward the input-tokens-per-minute limit. For team sizing, Anthropic suggests 200K–300K TPM per user for 1–5 users, falling to 10K–15K for 500+ users. For comparison with Codex's two-window system, see Codex usage limits explained.
How does usage tracking help avoid rate limits?
The best prevention is measuring before the limit, not recovering after it. Once you know your weekly baseline, you can spread your heaviest tasks over the days left before your reset.
- 1
Establish a baseline with two weeks of data
Check your daily totals for two weeks. Note which days and which kinds of tasks drove the most cache reads and output. Large refactors, long-lived sessions and parallel subagents are usually the big ones. - 2
Check /usage before expensive tasks
Before a big refactor or a workflow fan-out, look at both bars. On Pro, Max and Ultra, the same screen lists what counted against your limits over the last 24 hours or 7 days, by skill, subagent, plugin and MCP server. - 3
Shrink the context before heavy work
/compactat a natural break, or/clearbetween unrelated tasks. Every request carries the conversation, so a smaller context makes every later request lighter. - 4
Decide your plan for hitting the wall
Claude Code v2.1.234 and later can wait and continue the task automatically after a reset (/rate-limit-options). Or turn on usage credits with/usage-creditsto keep going at standard API rates, with a monthly spend cap you set.
session window, from your first message
fixed reset time per account
API-key rate limit, not plan quota
Everything Claude Code sends counts
File reads, tool calls, web searches, subagents and compaction all count against your windows. Anthropic's help center lists the length and complexity of your conversations and the effort level as factors. So a long context and/effort max on routine work are the fastest ways to burn a window. ⏳If you've already hit your limit and want to understand what you burned, see the recovery guide: Claude Code usage limit reached — what to do. If you also use Codex CLI, the Codex CLI usage check guide covers its own 5-hour and weekly windows.
Knowing your yardstick turns limits into a planning tool
After a week of comparing/usage percentages with your daily totals, limits stop being surprises and become a known constraint. You can plan around a known constraint, like disk space or memory.Sources checked November 8, 2026: Usage and length limits, Use Claude Code with your Pro or Max plan, What is the Ultra plan?, weekly limits promotion, Claude Code error reference, status line, manage costs, API rate limits, claude.com/pricing, TechCrunch on the 2025 weekly limits.
Related guides
Claude Code Usage Limit Reached? Here's What to Do
Locked out mid-session? See what you used and when the limit resets.
Codex Usage Limits: 5-Hour and Weekly Caps
Where to check Codex's 5-hour and weekly limits—and what local token history cannot tell you.
How to Check Claude Code Token Usage
See your Claude Code tokens by day and model, and how the built-in /usage compares.