AI Guides › Step-by-step guides
By Nigel Guy · 8 min read
Most people treat the usage limit as a fixed number of messages, then keep one enormous chat open for days because it "knows everything". That feels efficient, and it is the most expensive way to use Claude. Anthropic's help centre lists conversation length among the things that count against your limit, and Claude Code's docs say plainly that it sends your full conversation with every request. The limit is mostly a measure of how much context you carry, not how many questions you ask.
The rule: keep each conversation small and single-purpose, and carry the thread between conversations in a short handover note you write, not in a long history Claude has to re-read every turn.
You need a paid Claude plan to see the usage bars. At time of writing, Anthropic's pricing page lists Pro at $20 a month billed monthly ($17 a month on annual billing), about £15 at time of writing, and Max from $100 a month, about £75 at time of writing. Check the £ price at checkout, as it depends on exchange rate and tax. Max lets you choose 5x or 20x the usage of Pro.
How limits work, per the pricing page and help centre:
| Limit | What it does |
|---|---|
| Session limit | Resets on a rolling five-hour window |
| Weekly limit | Applies on paid plans, on its own schedule |
| Shared pool | claude.ai, Claude Code and Claude Desktop all count toward the same usage |
There is no fixed message count. What you spend depends on conversation length, attachments, tools such as research and web search, artifact creation, your model choice and your effort level. Exact allowances are not published as figures, so this guide does not quote any.
On claude.ai go to Settings > Usage. You will see progress bars for your five-hour session and your weekly usage. Glance at them at the start and end of a work block. In Claude Code, run /usage: on Pro and Max it shows plan usage bars and a breakdown of what is eating them, with behaviour flags such as long context or cache misses once one accounts for 10% or more of recent usage. Press d or w to switch between the last 24 hours and the last 7 days. These breakdown figures come from local session history, so usage from claude.ai or other devices is not included.
A new topic in an old chat means every reply drags the old topic along. Open a fresh chat when the task changes, and keep one chat per deliverable. In Claude Code, use /clear between unrelated tasks. The docs suggest running /rename first so you can find the session later with /resume.
This is the part that stops a fresh start from costing you the thread. Before closing a chat you want to continue, paste this in and save the answer somewhere you control.
You are helping me close a working session so I can continue it in a new chat.
Context: we have been working on [PROJECT_OR_TASK]. My goal is [END_GOAL]. The next thing I need is [NEXT_STEP].
Write a handover note I can paste into a new conversation. Include, in this order:
1. Goal and what "done" looks like, in two sentences.
2. Decisions already made, one line each, with the reason.
3. Facts and constraints that must not be forgotten (names, numbers, tone, deadlines) that appear in this chat.
4. What is finished, and where the latest version of any draft is (include the final text only if it is under [MAX_WORDS] words).
5. Open questions and the next three actions.
Rules: maximum [MAX_WORDS] words. Use only what is in this conversation; do not invent details. If something I would need is missing or unclear, list it under "Questions for me" instead of guessing.
Before you answer, check that someone with no access to this chat could carry on from your note alone.
Fill in the project, goal, next step and a word cap (300 is a sensible start). Read the note once and correct it, because it becomes the new chat's entire memory.
Open a new chat, paste the note, and add one line saying what you want first. If the work uses reference documents, put them in a Project. The help centre says cached portions of project documents count less against your limits than new content, so repeatedly uploading the same files is the costlier habit. Paid plans can also search past conversations and use memory, which saves re-explaining yourself.
Anthropic lists model and effort level as usage factors. Use a lighter model or lower effort for rewording, summaries and formatting, and save the heaviest settings for work that needs reasoning. Claude Code's docs recommend Sonnet for most tasks and Opus for complex architecture or multi-step reasoning. Switch with /model, and lower effort with /effort. Note that on some current models you cannot turn thinking off entirely, so effort is the lever.
Combine related questions into one message, and edit your message before sending so Claude does not need a clarifying round trip. A specific request ("add input validation to the login function in auth.ts") avoids the broad scanning that a vague one ("improve this codebase") triggers.
/context to see what is occupying space, and /mcp to disable servers you are not using./compact with instructions, for example /compact Focus on code samples and API usage, when you want to continue the same session. Compacting a large context is itself a large request, so /clear is cheaper when you do not need continuity.On a subscription the prompt cache lifetime is an hour, per the Claude Code docs. Your first message after a longer break reprocesses the whole context, so a big old session resumed cold is expensive. On Pro and Max, Claude Code offers to resume from a summary instead. Take that offer, or use your handover note and start fresh.
You have three honest options: wait for the reset time shown in the message, move up a plan, or turn on usage credits. Credits (Pro, Max 5x and Max 20x, web only) are enabled under Settings > Usage in the Usage credits section, are billed at standard API rates, need prepaid funds, and let you set a monthly cap. Set a cap before you need it. In Claude Code v2.1.234 or later, signed in with a subscription, the session can wait and continue the task on its own after the reset; cancel with Esc, or manage it through /rate-limit-options. A weekly reset can be days away, and Claude Code does not start that wait on its own when the reset is more than 24 hours out. Switching models with /model does not lift a session or weekly limit, which is shared across models.
/usage no longer flags long context or cache misses as a large share.