Claude usage limits are account allowances. They are not token counts, context windows, API bills, or subscription prices.
That distinction matters. A smaller request can preserve an allowance without changing your monthly subscription price.
Four terms that users often mix together
| Term | What it measures | What changes it |
|---|---|---|
| Tokens | Text and other model input or output | Context, tool results, thinking, and response length |
| Cost | A token-based charge or estimate | Model, token type, rate, and billing route |
| Usage limit | How much plan activity remains | Plan, model, effort, context, tools, and task complexity |
| Subscription | A recurring plan price | Your selected plan and billing terms |
Tokens can affect cost and plan usage. They do not make these measures identical.
Anthropic also separates usage limits from length limits. A usage limit applies across conversations during a time period. A length limit applies to one conversation’s context window.
The five-hour session limit
Anthropic says paid plans have a session limit that resets every five hours. The allowance varies by plan.
The meter does not count a fixed number of messages. Message length, current conversation length, tools, model choice, effort, and task complexity all matter.
Check Settings > Usage for the current meter and reset time on your account. Do not rely on an old message estimate from another account.
Weekly limits
Paid plans also have weekly limits. Anthropic shows the current weekly meters and reset times in Settings > Usage.
Pro and Max plans share these limits across Claude and Claude Code. Anthropic also says claude.ai, Claude Desktop, and Claude Code use the same allowance.
A separate model limit can also appear. Use the account meter to identify the limit you reached before you change models or plans.
Why a short request can use substantial allowance
A short prompt can sit on top of a large context. Claude Code sends conversation history and relevant tool results with later requests.
The main drivers are:
- a long conversation;
- large files or attachments;
- repeated tool results;
- higher effort or extended thinking;
- a more expensive model;
- subagents, teams, or scheduled work;
- a prompt cache miss after inactivity.
Prompt caching can reduce repeated processing costs. Cached content still affects usage, and cache behavior can change after inactivity or context changes.
What the Claude Code meters show
Use /usage to inspect session tokens and plan usage. Current versions can show
attribution for skills, subagents, plugins, and MCP servers.
The command can also flag long context or cache misses. Its local dollar figure is an estimate at configured rates.
For Pro and Max subscriptions, included usage does not create a matching API invoice. Check the applicable billing page for authoritative charges.
What happens at a limit
When an included allowance ends, you can wait for its reset. Eligible plans can also use usage credits or move to a plan with more capacity.
Usage credits and Claude API credits require separate attention. Anthropic says API rates are distinct from Pro and Max subscription pricing.
Review auto-reload and spend settings before you use API billing. This prevents an unexpected transition from included usage to metered spend.
How to make an allowance last longer
Start with the largest documented causes:
- Use
/clearwhen the task changes. - Use
/compactonly when the task needs continuity. - Keep
CLAUDE.mdfocused on essential instructions. - Use Sonnet for routine work when it meets your quality target.
- Lower effort for simple tasks.
- Disable unused MCP servers.
- Filter large command output before it enters context.
- Stop unused teammates and scheduled work.
These actions can reduce token use. They do not guarantee a fixed percentage or an identical result.
Where Halv fits
Halv processes context on your device before supported coding agent requests leave the machine. It can remove repeated or stale material and reports its measured token changes.
The result varies by task. A request with little removable context can show a small change or no change.
Our published Codex SWE-rebench checkpoint measured workflow cost, not Claude savings. It covered 14 tasks with three repetitions each.
Both arms passed 25 of 42 runs. Halv recorded $144.67 in model cost, compared with $337.50 for vanilla Codex.
The Halv workflow used 3.49 times as many tokens. That result does not establish Claude savings, universal token reduction, or equivalent results on every task.
Start a 7-day free trial. Halv Unlimited costs US$10 per month after the trial. A credit card is required. Cancel during the trial to pay nothing.
Sources
For a practical workflow, read how to reduce Claude usage. It covers shared usage, Claude Code tasks, and routing with JEV.
Official sources checked on October 14, 2026:
- How usage and length limits work
- Usage limit best practices
- Claude Code with Pro or Max
- Manage Claude Code costs
For a broader product comparison, read Claude Code vs Codex.