Table of Contents
- What are Claude Code rate limits and who sets them?
- How does the Claude Code 5-hour limit actually work?
- When does the Claude weekly limit reset, and does it reset for everyone at once?
- How much Claude Code usage do you get on Pro vs Max 5x vs Max 20x?
- Do Claude Code and the Claude.ai chat app share the same limits?
- Is the Anthropic API limited the same way as a subscription?
- Why did you hit your Claude Code limit so fast?
- How do you check how much Claude Code quota you have left?
- What happens when you hit the limit, and can you keep working?
- Should a heavy user move from Pro to Max or switch to API billing?
- How do Claude Code limits compare with ChatGPT and Codex for the price?
- How did we compare these options?
- How do you get started with Agent 37?
- What do people also ask?
Do not index
Claude Code rate limits are two stacked caps set by Anthropic: a 5-hour rolling session window and a 7-day weekly cap, both sized by your plan (Pro, Max 5x, or Max 20x), while API keys are limited by requests and tokens per minute instead of by the clock. If those limits keep interrupting long jobs, we recommend Agent 37, a managed hosting platform that runs Claude Code on a persistent, always-on instance, with managed OpenClaw and Hermes agents from $3.99 per month, so a paused task resumes the moment your window resets instead of waiting for you to reopen a laptop. Agent 37 is not a way around Anthropic's limits and it is not a model: your API key or subscription goes straight to Anthropic and keeps exactly the quota Anthropic gives it.
Key takeaways
- Two limit systems exist: subscription plans get a 5-hour session window plus a rolling 7-day cap, while API keys get requests-per-minute and tokens-per-minute ceilings that rise with usage tier and are billed per token.
- The 5-hour window starts at your first message, not on a fixed schedule, and the weekly window likewise runs from your own usage, so resets are personal rather than global.
- Anthropic's published estimates give Pro 10 to 40 Claude Code prompts per 5-hour window, Max 5x 50 to 200, and Max 20x 200 to 800, with separate weekly caps for Opus on Max plans.
- Claude Code, claude.ai, and Claude's desktop apps share one pool on Pro and Max, so a long browser chat eats the same quota as a coding session.
- Agent 37 keeps the work running between resets: Claude Code is provisioned on a persistent instance through one API call, your Anthropic key goes directly to Anthropic, and a built-in task board schedules recurring jobs.

What are Claude Code rate limits and who sets them?
Claude Code rate limits are the usage caps Anthropic applies to Claude Code, its terminal coding agent. Subscription users get a 5-hour rolling session window plus a 7-day weekly cap, sized by plan. API users are limited by requests and tokens per minute across usage tiers, and they pay per token instead of a flat monthly fee.
The confusion in most Reddit threads about Claude Code rate limits comes from mixing these two systems. A Pro subscriber at $20 per month is buying a fixed allowance that refills on a timer. An API customer is buying tokens, and Anthropic only throttles how fast those tokens can be consumed in any given minute.
Both systems are documented by Anthropic. The subscription windows are described in the Claude help center, and the API tiers are published on the Anthropic rate limits page. Anthropic changes the numbers without notice, so treat every figure below as a snapshot to confirm, not a contract.
How does the Claude Code 5-hour limit actually work?
The 5-hour window starts the moment you send your first message, not at a fixed clock time. Every prompt, tool call, and token Claude Code processes during those five hours counts against the window. When the five hours end, the counter resets fully, and the next message you send starts a fresh window.
In practice this means a session that starts at 8:10am resets at 1:10pm. If you hit the cap at 11am, you wait until 1:10pm, not until the top of the next hour. Claude Code prints the exact reset time when it blocks you, so note it before walking away.
The window counts consumption, not prompts. One prompt that triggers forty file reads and a large diff costs far more than a dozen one-line questions. That is why two people on the same plan report wildly different message counts per window.
When does the Claude weekly limit reset, and does it reset for everyone at once?
Anthropic's weekly limit is a rolling 7-day cap that started on August 28, 2025. It does not reset for everyone at the same moment: your seven days run from your own usage, and the reset time is shown in Claude Code and in Settings. Max plans also carry a separate weekly cap for Opus.
Anthropic introduced the weekly cap after reporting that a small group of accounts were running Claude Code continuously in the background, and it said the cap would affect fewer than 5% of subscribers based on usage at the time. The weekly cap sits on top of the 5-hour window, so you can be well inside your session allowance and still be blocked until the weekly reset.
There are two weekly numbers on Max plans: an all-models cap and a smaller Opus-only cap. Running Opus for every task is the fastest way to exhaust the second one while the first still shows plenty of room.
How much Claude Code usage do you get on Pro vs Max 5x vs Max 20x?
Claude Code is not included on the Free plan. Anthropic's own estimates put Pro at 10 to 40 Claude Code prompts per 5-hour window, Max 5x at 50 to 200, and Max 20x at 200 to 800. Weekly, Pro gets an estimated 40 to 80 hours of Sonnet, Max 5x 140 to 280, and Max 20x 240 to 480.
Plan | Published price | Claude Code per 5-hour window (Anthropic estimate) | Weekly Sonnet (Anthropic estimate) | Weekly Opus cap |
Free | $0 | No Claude Code access | Not applicable | Not applicable |
Pro | $20 per month | 10 to 40 prompts | 40 to 80 hours | Shares the general cap |
Max 5x | $100 per month | 50 to 200 prompts | 140 to 280 hours | 15 to 35 hours |
Max 20x | $200 per month | 200 to 800 prompts | 240 to 480 hours | 24 to 40 hours |
Prices come from the Anthropic pricing page; confirm current figures there, since Anthropic has adjusted both prices and allowances more than once. The ranges are wide on purpose: Anthropic measures them in Sonnet-equivalent work, and codebase size, model choice, and tool use move a real session anywhere inside the band.
Do Claude Code and the Claude.ai chat app share the same limits?
Yes. On Pro and Max plans, Claude Code, the Claude.ai web and desktop apps, and newer surfaces such as Claude Cowork draw from one shared pool. A long chat in the browser at 9am leaves less for a coding session at 11am. Team premium seats follow the same per-seat model.
The shared pool is why "Claude Pro limits too low" complaints often come from people who never opened Claude Code that day. Uploading a 200-page PDF to claude.ai and asking six questions about it can consume a meaningful slice of a Pro window before any code is written.
If you want hard separation between chat and coding, the only clean way is to run Claude Code on an API key, which has its own meter and its own bill. We cover when that trade is worth it below.

Is the Anthropic API limited the same way as a subscription?
No. API keys have no 5-hour or weekly window. Anthropic throttles them by requests per minute, input tokens per minute, and output tokens per minute, with ceilings that rise through usage tiers as you spend more. You pay per token, so the practical limit is your budget, not a clock.
Subscription plans (Pro, Max) | API keys (pay per token) | |
What is capped | Total work per 5-hour window and per 7 days | Requests per minute, input tokens per minute, output tokens per minute |
How it refills | Timer: 5-hour window and rolling week | Continuously, as a token bucket, every second |
How it grows | Upgrade plan (Pro to Max 5x to Max 20x) | Automatically through usage tiers as cumulative spend rises; workspace overrides on request |
What you see when blocked | "Session limit reached" style message with reset time | HTTP 429 ratelimiterror with a retry-after header |
How you pay | Flat monthly fee | Per million tokens, for example Sonnet at $3 input and $15 output at publication |
Best for | Interactive daily coding at a fixed cost | Automation, parallel sessions, and anything unattended |
The API uses a token bucket, so a short burst can trip a 429 even when your average throughput is low. Anthropic documents the exact per-tier numbers on its rate limits page, including monthly spend ceilings per tier that are separate from the per-minute throttles.
Why did you hit your Claude Code limit so fast?
Quota drains faster when each turn is expensive. Opus burns quota faster than Sonnet, every message resends the whole context, subagents multiply calls, and long sessions carry a bloated history. The same ten prompts can cost ten times more on one day than another depending on model and context.
Context is the quiet culprit. Claude Code sends the full conversation history, your CLAUDE.md, and every file it has read with each new turn. A session that has read 30 files is paying for those 30 files on every prompt until you compact or clear it.
Subagents and parallel tool calls are the second multiplier. One prompt that fans out to four subagents can cost five full model calls, each with its own context. Through late 2025 and into 2026, users also reported quota draining with no matching work; Anthropic acknowledged and fixed accounting bugs, so check the Anthropic status page before you blame your own habits.
Seven tactics that stretch a window:
- Set Sonnet as the default model and switch to Opus only for planning or a hard bug.
- Run /compact when a session passes a few dozen turns, and /clear when you change tasks.
- Ask for targeted reads (a function, a file section) instead of entire directories.
- Keep CLAUDE.md short; it rides along on every prompt.
- Limit subagents to tasks that truly parallelize, such as independent test suites.
- Batch small chores into one prompt before a reset rather than ten prompts after it.
- Move unattended or scheduled jobs to an API key so they never touch your session pool.
How do you check how much Claude Code quota you have left?
In Claude Code, type /usage to see the percentage of your 5-hour and weekly allowances used and when each resets; /status shows plan and model. On claude.ai, open Settings and then Usage. For API keys, read the anthropic-ratelimit headers on every response, which report the limit, remaining amount, and reset time.
API users running Claude Code can also type /cost to see the dollar total for the current session. Get in the habit of checking /usage before starting a large refactor, because the weekly number is the one that surprises people on a Thursday afternoon.
What happens when you hit the limit, and can you keep working?
Claude Code stops accepting new prompts and prints a message such as session limit reached with the reset time; API calls return HTTP 429 with a retry-after header. You can wait, switch to a cheaper model, enable extra usage where Anthropic offers it, or point Claude Code at an API key and pay per token.
Anthropic added an extra usage option that bills overflow at API rates once a plan's included allowance is spent; it is switched on from plan settings and availability by plan has changed over time, so check your own account. The fallback that always works is exporting ANTHROPICAPIKEY and restarting Claude Code, which moves the session to token billing immediately.
The part limits cannot fix is the laptop. If a 90-minute migration stops at 11am and resets at 1:10pm, the job only resumes if something is still running at 1:10pm. That is the gap we built Agent 37 to close: Claude Code hosting on a persistent instance keeps the repo, sessions, and caches alive, and the task board can schedule the retry for the reset time while you are in a meeting. Agent 37 does not add quota; it keeps the worker alive so the quota you already have is never wasted on a dead session.

Should a heavy user move from Pro to Max or switch to API billing?
Move from Pro to Max 5x when you hit the weekly cap in more than half your working weeks, since the $80 difference buys five times the allowance. Switch to API billing when you need unattended automation, parallel sessions, or predictable throughput, because tokens are metered continuously and never lock you out for five hours.
The rough math: at Sonnet's published $3 per million input tokens and $15 per million output tokens, the $80 gap between Pro and Max 5x buys about 5 million output tokens or more than 25 million input tokens on the API. A developer doing a few interactive sessions a day almost never consumes that, so Max 5x is usually the better deal for humans at a keyboard. An agent running overnight on a schedule is the opposite case: it needs the API, because a subscription window will block it halfway through.
Trust is the other variable. Stack Overflow's 2025 developer survey found that 84% of developers use or plan to use AI tools while 46% distrust the accuracy of the output, which argues for keeping a human review step rather than buying the largest plan and letting the agent run unsupervised. Agent 37's task board exists for exactly that review step: work is assigned, outputs are reviewed, and recurring jobs are scheduled from one screen. For teams provisioning many workers, the Agent 37 Cloud API hands each user their own always-on instance behind one API call, and the Agent 37 pricing page shows managed hosting from $3.99 per month on a bring-your-own API keys model, so the hosting cost is clear before the Anthropic bill is.
How do Claude Code limits compare with ChatGPT and Codex for the price?
Both vendors use the same shape: a $20 plan with a 5-hour window and a weekly cap, and $200 top tiers. OpenAI's Codex limits on ChatGPT Plus and Claude Code on Claude Pro are both sized for a few focused sessions a day. Neither $20 plan is built for all-day agent automation; API billing is.
OpenAI publishes its Codex allowances in the OpenAI help center and adjusts them as often as Anthropic does, so compare current pages rather than screenshots from Reddit. The structural difference is in the middle tier: Anthropic sells Max 5x at $100, while ChatGPT jumps from Plus to Pro at $200, so a user who needs roughly five times a Pro allowance pays less at Anthropic.
Whichever vendor you pick, the hidden cost is the always-on computer. An always-on worker on the smallest sensible hyperscaler VM lands near $25 a month before storage, schedulers, and networking, which is the gap managed AI agent hosting from $3.99 per month closes for OpenClaw and Hermes, with Claude Code, Codex, and Pi provisioned on the same kind of persistent instance through one API call.
How did we compare these options?
We compared Anthropic's published plan pages, the Claude Code and API rate limit documentation, Anthropic's August 2025 weekly limit announcement, and OpenAI's Codex usage page, then checked the figures against what Claude Code reports through /usage. Plan capacities are Anthropic's own estimates, not our measurements, and vendors change them without notice.
Agent 37 details come from our own product pages and documentation. We did not run a controlled benchmark of quota consumption, and we recommend confirming every price on the vendor's site before choosing a plan.
How do you get started with Agent 37?
Create an Agent 37 account, deploy a managed OpenClaw or Hermes instance in one click from $3.99 per month, or provision Claude Code on a persistent instance through one API call. Your Anthropic key goes directly to Anthropic, your data stays in an isolated container, and the task board schedules recurring jobs around your resets.
The browser-based terminal, file browser, and live Linux desktop let you check on a running job from a phone, and 1000+ integrations via Composio connect the agent to Gmail, Slack, GitHub, and WhatsApp without extra setup. Start with the Agent 37 quickstart and have a persistent instance running before your next 5-hour window resets.
What do people also ask?
How does the Claude Code 5-hour limit actually work and when does it reset?
The window opens at your first message and closes exactly five hours later, when the full allowance returns. Everything Claude Code does in that window counts, including file reads and tool calls, so heavy prompts use more than light ones. Claude Code prints the exact reset time when it blocks you, and /usage shows it at any time.
When does the Claude weekly limit reset, and does it reset for everyone at the same time?
No. The weekly cap is a rolling 7-day window tied to your own usage, so your reset time differs from other users'. Run /usage in Claude Code or open Settings and then Usage on claude.ai to see your personal reset time. Max plans have an additional Opus-only weekly cap that can run out before the general cap.
How much Claude Code usage do I get on Pro vs Max 5x vs Max 20x?
Anthropic's published estimates are 10 to 40 Claude Code prompts per 5-hour window on Pro ($20 per month), 50 to 200 on Max 5x ($100), and 200 to 800 on Max 20x ($200), with weekly Sonnet estimates of 40 to 80, 140 to 280, and 240 to 480 hours respectively. The Free plan does not include Claude Code. Confirm current figures on anthropic.com/pricing.
Can I keep working after hitting the Claude Code limit?
Yes, in four ways: wait for the reset, switch to a cheaper model such as Sonnet or Haiku for the rest of the window, turn on Anthropic's extra usage option where your plan offers it, or export an ANTHROPICAPIKEY and restart Claude Code to bill per token with no session window. Hosting Claude Code on a persistent Agent 37 instance lets a paused job resume automatically at the reset time.
Do Claude Code and the Claude.ai chat app share the same limits?
Yes. On Pro and Max plans, Claude Code, the claude.ai web app, the desktop apps, and Claude Cowork all draw from one shared pool per 5-hour window and per week. A long chat with large uploads reduces what is left for coding. Only an API key gives Claude Code a separate meter and a separate bill.
