Articles about Claude Pro limits explain the mechanics well. What they leave out is what to do at the moment you hit the limit. The titles at the top of search results already describe that moment: “Hit the limit with a lot left to do?” and “Why can't I use Claude even with a paid subscription?” Yet most answers stop at “wait or upgrade.” This article adds a third option and compares all three on the same basis: cost.
How it works: limits count tokens, not messages
Limits use rolling windows, not a fixed monthly allocation. The window starts with your first message and capacity is restored after a set period. There is also a separate weekly cap, so heavy use over a short period can hit the weekly limit before the 5-hour window limit.
The important point is that consumption is measured in tokens, not messages. Each exchange resends the entire conversation history, so the same “message” consumes more as the conversation gets longer. File-reading and editing use can reach tens of thousands of tokens per step. That is almost always why you “hit the limit quickly”; it is not a setting.
Common misconception: upgrading does not open a window that is closed now
Changing to a higher plan increases the limit starting with the next window. It does not open the window that is currently closed. If you hit the limit in the middle of deadline-driven work, upgrading will not solve it; making the wrong call here can increase your bill while leaving that day's work stalled.
The reset time is not fixed either. The window starts when each person sends their first message, so it varies from person to person. The resume time shown on the limit screen is the only accurate information.
Compare the three options by cost
| Option | Does it work immediately? | Best for |
|---|---|---|
| Wait for the window to reopen | No (wait) | No deadline; you can put it off for a few hours |
| Switch to a higher plan | No (starting with the next window) | The limit is reached every day; capacity is consistently insufficient |
| Switch that task to pay-as-you-go | Yes (immediately) | It must be finished today; it only happens a few times a month |
# 한도에 걸린 작업만 종량제로 넘기는 두 줄.
# 구독은 그대로 둡니다 — 변수가 있는 동안만 키로 청구되고, 지우면 돌아갑니다.
export ANTHROPIC_BASE_URL=https://api.kunavo.com
export ANTHROPIC_AUTH_TOKEN=sk-kn-...
# 클로드 코드의 기본 모델과 opus·sonnet 별칭은 최신 모델을 따라가므로,
# Kunavo가 아직 제공하지 않아도 404가 나지 않도록 모델을 고정합니다.
# sonnet 별칭이 부르는 Sonnet 5.5는 Kunavo에 없어서, 고정하지 않으면
# /model sonnet, opusplan의 실행 단계, sonnet 서브에이전트가 404를 받습니다.
# opus 별칭은 Opus 5.5(claude-opus-5-5)로 고정 — 클로드 코드 v2.1.280 이상 필요(이전 버전은 claude update).
export ANTHROPIC_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5-5
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-5
# 보조적인 내부 호출은 가장 싼 모델로
export ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-4-5How much does switching to pay-as-you-go actually cost?
Rates are taken directly from the catalog. Per 1M tokens:
| Model | Input / output |
|---|---|
| Claude Haiku 4.5 | $0.70 / $3.50 |
| Claude Sonnet 5 | $1.40 / $7.00 |
| Claude Opus 5.5 | $2.80 / $14.00 |
In terms of units of work, one step with 25,000 input tokens / 1,200 output tokens costs approximately $0.0434 based on Claude Sonnet 5 — about 23 steps for $1. A task that takes 20 steps to complete costs approximately $0.868. Since the cost is deducted from your prepaid balance, you are not billed in months when you do not use it. Break-even calculations are covered in detail in Claude Code Pricing. If payments with Korean cards are blocked, see Claude API Pricing and Payments.
Three ways to hit the limit less often
- Break up conversations. Starting a new conversation between unrelated tasks alone can greatly reduce consumption. An endless thread is the most expensive option.
- Choose the model according to the size of the problem. Do not make the largest model the default — in the table above, the output rate for Claude Opus 5.5 is approximately 4 times that of Claude Haiku 4.5.
- Let caching do its work. For usage that sends the same context each time, it can greatly reduce input costs (prompt caching).
For Claude Code operations (CLAUDE.md, permissions, custom commands), see Claude Code usage guide. The detailed English version is Claude Pro and Max limits.
Frequently asked questions
How are Claude Pro limits structured?
They use rolling windows, not a fixed monthly allocation. The window starts when you send your first message, and capacity is restored after a set period. There is also a separate weekly cap, so heavy use over a short period can hit the weekly limit before the 5-hour window limit. Consumption is measured in tokens, not messages, so a long conversation or large attachment counts more even though it is still a single “message.”
If I upgrade my plan, can I use it right away?
No. This is the most common misunderstanding. Changing plans increases the limit for the next window; it does not open a window that is currently closed. If you hit the limit while working on a deadline, upgrading will not solve the problem. The only ways to continue right then are to wait or switch that task to a pay-as-you-go API key.
Why do I hit the limit so quickly?
Usually, because the conversation is long. Each exchange resends the entire history so far, so the longer the conversation gets, the more each message consumes. Agentic use—having it read and edit files—is heavier and can use tens of thousands of tokens in one step. Starting a new conversation with only the key points carried over works more reliably than changing any setting.
How can I continue without waiting after hitting the limit?
You can switch just that task to an API key; there is no need to cancel your subscription. Tokens are billed by usage only while ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN are set, and clearing the variables switches you back to the subscription. As a benchmark, one step with 25,000 input / 1,200 output tokens costs about $Claude Sonnet 5 on 0.0434—about 23 steps for $1.
Can I check when the limit resets?
The window starts when each person sends their first message, not at a time shared by all users. So the reset time varies from person to person, and there is no fixed answer to “what time will the limit be lifted?” The screen shown when you reach the limit displays when you can use it again; that is the only accurate information.
Is pay-as-you-go more expensive?
It depends on your usage. The calculation is one division: monthly fee ÷ cost per task = break-even number of tasks. One 20-step task costs about $0.868, so a subscription is cheaper in months when you run more tasks than this, and pay-as-you-go is cheaper in months when you run fewer. You can also use a subscription normally and switch to a key only on days when you hit the limit.