A lot of it comes down to one habit: the length of your current conversation counts against your usage. So the longer a chat gets, the more each reply costs, and one endless thread quietly burns through your limit while a fresh one barely touches it. Fix that habit and the wall moves a long way back.
You hit your AI limit fast because the length of your current conversation counts against your usage, so long threads cost more and more per reply. Start a new chat for each new topic, keep reference material in a Project rather than pasting it into every chat, and use a lighter model for simple jobs.
Claude charges less for decks, docs and designs until 15 October
- What you get. When Claude creates or edits a document, slide deck or design in a chat, the next 10 messages in that chat use 50% less of your five-hour limit, and the count restarts each time Claude creates or edits another one. A very long reply is only partly discounted.
- Start it from the Output menu. Open a new chat and choose Document, Presentation or Design from the Output menu, and the discount covers your first message too. Ask for one partway through an ordinary chat and the message that makes it counts at the full rate.
- What it leaves alone. Your weekly limit counts as normal, and Claude Code and anything paid for with usage credits (the extra usage you can buy) are left out.
- Who gets it, and when it ends. It applies automatically on Pro, Max and Team plans, and on the Free plan since 7 October, until 11:59pm Pacific Time on 15 October 2026, says Anthropic’s terms (read 9 October 2026), and Anthropic may end it early.
What changed on 22 September 2026
- You have a saved reset, and it expires. Pro, Max and Team plans were given one reset, which sets either your five-hour or your weekly limit back to full. Anthropic’s developer account says to use it any time until 22 October. It sits under Settings > Usage, in the Resets section, as a button marked “Reset for free”.
- Use it on a computer, and only when you are stuck. Anthropic’s help page (read 29 September 2026) says the button is missing from the phone app, that a reset cannot be undone, and that it is lost if you downgrade or cancel first.
- The five-hour limit went up. Anthropic’s announcement covers Pro, Max, Team and seat-based Enterprise plans and gives no figure. Its developer account puts the rise at 20% for Claude Code, and neither says whether it is permanent.
- The new default model stretches your limit. Opus 5.5 is now the default on paid plans. Anthropic’s developer account, in a post on X, says it is priced lower than Opus 5, “so your 5-hour and weekly limits go 25% further”.
- Your bill is the same. The price cut is for developers who pay per use, and Anthropic announced no change to plan prices.
What a limit actually is
Your plan gives you a budget of computing work over a window of time, and each action uses a different amount of it.
A quick question on a light model is cheap. A long thread, a big file, or the heaviest reasoning model is expensive. You reach the limit when the work adds up, however few messages you have sent.
As of 29 September 2026, this is what each provider publishes.
Claude
- Short-term limit: resets every five hours, on a rolling basis.
- Weekly limit: one limit across all models, reset at a fixed time assigned to your account.
- Free against Pro: Anthropic says Pro gives at least five times the usage per session of the free plan.
- Where to check: Settings > Usage, on Pro, Max, Team and seat-based Enterprise. Its bars show how much of your five-hour session and weekly limit you have used, the time left in the session, and when the weekly limit resets, says Anthropic’s usage tips (read 6 October 2026).
- Fixed message count: none published, and Anthropic says it may adjust caps with demand.
ChatGPT
- Free and Go: unlimited everyday text chats, subject to safeguards that prevent abuse.
- Tools: file uploads, images, voice and data analysis have their own limits, and ChatGPT tells you when you reach one.
- Plus: may cap messages, especially during high demand.
- Fixed message count: none on OpenAI’s plan pages.
Why yours runs out so quickly
Anthropic’s list of what eats your usage includes message and file sizes, how long the current conversation is, tools like web search and research, and the model and effort level you pick. Three of those are habits you can change today.
- One endless chat. Because the length of the conversation counts against your usage, a chat you have kept going for days is one of the biggest drains, because every new reply has the whole chat behind it.
- Re-uploading the same file. A document you attach stays in the thread. Attach it again in the next message and you have paid for it twice.
- The heaviest model for trivial jobs. Using the top reasoning model to reword an email costs far more than a light model would, for no real gain.
How to make it last
1. Start a fresh chat for each new topic
This is the big one. A new chat carries none of the old history, so every reply is cheap again. When you finish a task and move to something unrelated, open a new conversation rather than carrying on in the old one. You lose nothing you need and you stretch your limit a long way.
2. Keep reference material in a Project
This is the tip most people miss, and it comes straight from Anthropic. Put the documents or instructions you keep reusing into a Project, and in their words:
"Content in projects is cached and counts less against your limits when reused."
So instead of pasting the same brief or file into every chat and paying for it in full each time, you load it once into the Project and it costs less each time you reuse it. The saving lapses after a long break, when your first message back counts that content in full again.
3. Batch your questions, and pick the right model
Ask related things in one message rather than firing off five in a row, since each message adds to the conversation length that counts against you.
Then match the model to the job: a light, fast model for quick rewrites and lookups, the heavy reasoning model only when the task actually needs it. If you are not sure which is which, see which AI model to use.
The paste-in that keeps chats lean
Drop this into a Project's instructions, or at the top of a chat you know will run long. It tells the AI to keep its answers tight and to nudge you when a thread is getting bloated, both of which save your budget.
You are my usage minder, the one who keeps my AI working all day on one budget instead of burning out by lunch. My limit is driven mostly by how long each conversation gets and how heavy the work is, so your job is to keep every reply lean and to warn me before a chat turns into a drain. SET UP ONCE (work these out from how I write, do not ask me): - How I work: I want the answer, not the working. Assume I am comfortable and busy. - What "done" looks like: the outcome, the decision, and anything I have to act on. Nothing spare. - What I mostly do here: read it from my questions, so you know what counts as heavy for me and where to spend the length budget. - My default format: prose, unless the answer is genuinely a list or a table. Match how I asked. YOUR JOBS 1. When my question is simple, keep the answer short. - Lead with the answer. A simple question gets one to three sentences of plain prose. - Skip the warm-up like "Great question", do not restate my question, and do not recap the steps you took. - Use a heading, table or list only when it carries real structure, never as decoration. 2. When the task is small, keep the effort small. - For something simple, a reword, a lookup, a quick fix, stay light and fast. - Save deep step-by-step reasoning for when I say "full version" or "think it through". 3. Handle my files without wasting the limit. - When I paste or attach a long document, use it and refer to it. Do not quote the whole thing back to me. - If I have already given you something in this chat, do not ask me to paste it again. - If I keep pasting or referring to the same document across chats, tell me once to load it into a Project so it costs me less each time. 4. Watch the length of the chat and warn me. - If this conversation has drifted off its first topic, say so and suggest I start a fresh chat. - If it is getting long, tell me in one line that a new chat resets the cost. - If I fire off several related questions, pull them into one answer rather than many. 5. Flag when I am over-powering the task. - If I am running a light job on the heaviest reasoning model, remind me in one line that a lighter, faster model does it for less. - Say it once per chat at most. 6. When I start a new task, size it before you begin. - In one line, tell me whether this is a light job (a rewrite, a lookup, a quick fix), a middling one, or a heavy one (long documents, many steps, careful reasoning). - Recommend a model and an effort level to match: the lighter, faster model for a light job, the strongest model and a higher effort only for a heavy one. You cannot change the model yourself, so tell me to switch it in the model menu if I should. - If you do not know which plan or models I have, ask me once and do not assume. - Never quote a figure for how much of my limit a model or a task uses. You cannot see my usage, so point me to my own usage screen for that. - If the new task has nothing to do with what we were doing, tell me to start it in a fresh chat. RULES - Never trade being correct for being short. Keep full error messages, warnings, numbers, and anything I need to act on safely. - If being brief would drop a caveat that changes what I should do, keep the caveat and cut something else instead. - When I say "full version" or "explain", set all of this aside and go as deep as the task needs. BEFORE YOU SEND, CHECK YOURSELF - Did I lead with the answer and cut the padding? - Did I keep every fact I actually need to act on? - Is this chat getting long or off topic, and did I flag it if so?
It works because a shorter reply is a cheaper reply, and because the AI is better placed than you are to notice when a thread has grown heavy.
The honest catch
The exact limits move. Providers raise and lower them with demand, change them by plan, and update the numbers without much fanfare, so any figure you read online may already be stale. Do not trust a hard count from a blog, including this one.
The one reliable move is to open your own usage screen, on Claude that is Settings > Usage on the paid plans (Pro, Max, Team and seat-based Enterprise), and watch how a long thread versus a fresh one actually moves the bars. That teaches you your real budget faster than any number.
Try it now
Next time you sit down with Claude or ChatGPT, do one thing: start a new chat for each separate task instead of one rolling conversation. If you have a document or brief you keep reusing, put it in a Project so it costs you less.
Questions people ask
What should I do if my Claude AI usage limit has reached?
Wait for the reset: the short-term limit resets every five hours on a rolling basis, and on paid plans Settings > Usage shows the time left. Then start a fresh chat for each new topic and keep reused material in a Project, so the next session lasts longer.
Why is AI usage limited?
Your plan gives you a budget of computing work over a window of time, and each action uses a different amount of it. A long thread, a big file or the heaviest reasoning model costs the most, so you reach the limit when the work adds up.
Why do I hit the limit so fast?
Because the length of your current conversation counts against your usage. The longer a chat gets, the more work each new reply costs, so one endless thread drains your usage far quicker than several short ones. Big file uploads and the heaviest reasoning model add to it too.
Does starting a new chat really help?
Yes. A fresh chat carries none of the old history, so each reply is cheap again. Start a new one whenever you move to a new topic, and you will get much more done before you reach your limit.
What are the usage limits on ChatGPT and Claude?
Claude's session usage resets every five hours, and its weekly cap resets at a fixed time for your account. On a Pro, Max or Team plan, check Settings, then Usage, for a saved reset. ChatGPT's free and Go tiers give unlimited everyday text chats, though uploads, images and other tools have their own limits.