Users who run long, repeated conversations with Claude — the kind of back-and-forth typical of research or experimentation — tend to hit usage limits first. This article rounds up the usage-saving methods described in Claude's official support documentation, covering both claude.ai chat and Claude Code (Anthropic's terminal-based coding tool), and also covers the Claude Code weekly limit change scheduled for September 14, 2026.
What actually determines how fast you use up Claude usage?
Claude usage isn't simply a count of messages — official support documentation states it depends on the length and complexity of your conversations, which model you're using, and the effort level (a setting that trades off speed and depth of reasoning) you've selected. Usage across claude.ai, Claude Code, and the Claude desktop app all draws from the same limit. Different plans (Free, Pro, Max, Team, and so on) come with different allowances, and you can check exactly how much you have left under Settings → Usage, where progress bars show both your five-hour session and weekly usage — seeing those two bars confirms you're looking at the right screen.
How does planning your conversations save usage?
Official usage guidance starts with a simple habit: think before you send. Firing off a vague question and then following up with a string of corrections uses far more usage overall than providing what's needed up front.
- Batch related questions into one message — combining several related questions in a single message drains your limit more slowly than sending each one separately.
- Review your prompt before sending — a well-edited prompt reduces the number of follow-up clarification exchanges.
- Use chat search and memory — on paid plans, you can pull relevant information from past conversations into a new chat instead of re-explaining context.
- Front-load context for coding and writing tasks — provide your full coding environment details in the first message, or send an entire piece of writing at once rather than in pieces.
How does the Projects feature cut usage on repeated work?
If you work with the same reference material across many conversations, Projects is the most effective way to save usage. Content uploaded to a project is cached (caching means content that's already been processed is stored and reused instead of being reprocessed), and according to official documentation, cached content doesn't count against your limits when reused.
- Upload core reference material to project knowledge upfront — so you don't need to re-attach it in every new chat.
- Projects use RAG — Retrieval-Augmented Generation (RAG) means Claude searches for and loads only the relevant parts of your documents instead of reading everything, which lets it handle larger amounts of information more efficiently.
- Keep project instructions concise — reserve broad context and key guidelines for the project instructions, and put task-specific instructions in the chat itself.
- Clean up unused files regularly — remove files from a project once you're no longer actively referencing them.
Does turning off extended thinking and tools actually reduce usage?
Yes — official documentation recommends turning off extended thinking (a mode where Claude reasons for longer before answering) when a task doesn't need it. The same logic applies to lowering the effort level for routine tasks, and to disabling tools and connectors (integrations with external services) you're not using, since official guidance notes these are token-intensive (a token is the basic unit of text an AI model processes).
- Reserve extended thinking for tasks that need it — many everyday requests don't require deep multi-step reasoning.
- Lower the effort level for simple, repetitive tasks — this uses less of your limit per message.
- Turn off tools and connectors you're not using — disable integrations that aren't relevant to your current task.
Caution: only turn these off when the task genuinely doesn't need them. Complex research or multi-step experiments that actually require deep reasoning can suffer in quality if you disable extended thinking across the board.
How do /clear and /compact help manage usage in Claude Code?
In Claude Code, managing your context window (the maximum amount of conversation and code Claude can hold in memory at once) is effectively how you manage usage. As a session runs longer, the context window fills with prior messages, files Claude has read, and command output. Official best-practice documentation recommends managing this directly with the following commands.
- /clear — resets the context window entirely when you move to an unrelated task. Using it often keeps sessions lighter.
- /compact — summarizes the conversation so far to free up space while keeping the thread of what you were working on. Adding an instruction, like /compact focus on the API changes, targets the summary at what matters.
- /context — shows exactly where your tokens are going: system prompt, tools, memory files, and conversation history.
- Subagents — asking Claude to "use a subagent to investigate X" runs research-heavy work, like reading many files, in a separate context window and returns only a summary, keeping your main conversation's context — and usage — lower. This is especially useful for exploratory, experiment-style work.
- /btw — for side questions unrelated to the current task; the answer never enters conversation history, so it doesn't grow your context.
Caution: /clear wipes your current context entirely. If you still need earlier context, consider /compact instead, or double-tap Esc to open the rewind menu and summarize only from a chosen point.
What's changing with Claude Code weekly limits on September 14, 2026?
The temporary 50% increase to standard Claude Code weekly usage limits, in effect from May 13 through September 13, 2026, is about to end, and the policy is changing after that. Official support documentation states only that once the promotion ends, weekly usage limits return to their standard levels; it covers Pro, Max, and Team plans plus legacy seat-based Enterprise plans, while Free plans and consumption-based Enterprise seats were never part of this promotion.
Separately, Anthropic's official X account announced that starting September 14, standard weekly limits will be permanently raised by 25%. In the same announcement, it also stated that compared to today — while the 50% boost is still active — this works out to a 17% reduction. In other words, the new standard is higher than it was before the promotion, but lower than what you're currently experiencing during the promotion. As of this writing, that 25%/17% figure appears only in the X announcement, not yet in the official support article, so it's worth checking the support page again once the change takes effect. This change concerns weekly limits only; the announcement doesn't mention the five-hour session limit.
If you're planning heavy research or repeated experiments, finishing them before the promotion ends on September 13 is a practical way to make the most of the current higher limit.
Frequently asked questions
Q. What happens when I hit my usage limit?
You'll need to wait until your current five-hour session or weekly cycle resets before you can use Claude again. You can check your exact remaining usage and reset timing under Settings → Usage.
Q. Is turning off extended thinking always a downside?
No. Official guidance recommends turning it off only for tasks that don't need it — for genuinely complex reasoning tasks, it should stay on.
Q. What's the difference between /compact and /clear in Claude Code?
/clear wipes the context window entirely and starts fresh, while /compact summarizes what's happened so far and just frees up space. Use /clear when switching to an unrelated task, and /compact when continuing the same task but needing more room.
Q. Are Claude Code limits going up or down after September 14?
Both, depending on your reference point. According to Anthropic's announcement, the new standard limit is 25% higher than it was before the promotion, but about 17% lower than what's currently available during the promotion period.
Next steps
Most of these changes can be applied right away with a settings tweak. For the full picture on how usage limits work, see the Claude usage limits guide, and for the latest on the September 14 change, see the Claude Code weekly limits update.