Claude Code usage limit reached: what to do, and how to hit it less
When Claude Code says you have hit your session limit or your weekly limit, you have used up a usage window on your plan, and switching models does not bring it back. The message shows when the window resets. Until then you can wait, turn on usage credits, or let Claude Code continue the task after the reset. To hit the limit less often, send less with every turn: each request carries the whole session again, and all of it counts.
Updated 2026-11-16
Which limit did you hit?#
The wording of the message tells you which ceiling you reached. They are not the same thing, and the fix is different for each. This table follows Anthropic’s cost docs.
| The message says | What it means | What helps |
|---|---|---|
| You’ve hit your session limit, or your weekly limit | A usage window on your subscription plan is used up. It is shared across all models. | Wait for the reset time in the message, or use usage credits. Switching models does not help. |
| You’ve hit your Opus limit, or your Sonnet limit | The limit for that one model family is used up. | Switch to a model outside that family with /model and keep working. |
| You’ve hit your spend limit | You are on usage credits, and they have reached a spend limit you or your admin set. | Raise the limit, or wait for the plan reset the message names. |
| A context or auto-compact warning | Not a usage limit. The conversation is close to the size where Claude Code summarizes older history. | Run /compact, or /clear if you do not need the history. |
On Team and Enterprise plans, Anthropic describes each seat’s allowance as a rolling five-hour window and a weekly window, shared with Claude chat and Cowork. So work you do in the Claude apps draws from the same allowance as Claude Code.
What to do right now#
- Run
/usage. On a subscription plan it shows your plan usage bars, so you can see which window is full and when it resets. - If you need to keep working, run
/usage-credits. Usage credits let you continue past the plan limit and are billed on top of your plan. On a team plan without billing access, the command sends a request to your admin. - If the work can wait, let it. Recent versions of Claude Code can wait for the reset and then continue the interrupted task on their own;
/rate-limit-optionsshows the choice. - If the message names one model, switch with
/model. That only works for the model-specific message, not for the session or weekly one.
Starting a new session does not restore a plan window. It does help the next window last longer, which is the rest of this guide.
Why the limit goes faster than you expect#
Claude Code sends your full conversation with every request, and each time Claude uses tools it sends another request with those results attached. A one-line question late in a long session still draws usage for the whole conversation behind it.
We asked the same 30 questions in Claude Code 2.1.325 on Claude Sonnet 5.5, three full sessions a side: once with a test memory in the layout Claude Code itself writes, as it was, and once after MinniMemoryMCP 1.6.0 sorted it, with its server and hook on. In the sessions before any change, the tokens split like this:
| Token kind | Tokens | Share of all tokens |
|---|---|---|
| Fresh input | 490 | under 1% |
| Cache write | 134,041 | 3% |
| Cache read | 4,133,084 | 96% |
| Output | 46,520 | 1% |
Almost none of it was new text. Most of it was the same session being read back, turn after turn. Anthropic lists the usual reasons a long session climbs:
- Long context. Everything still in the conversation is sent again on every request.
- Cache misses. The first message after a long break misses the prompt cache and processes the whole context again.
- Scheduled tasks and loops. They fire on their interval even while you are away, and each run sends the full context.
- Subagents and agent teams. Each one sends its own requests on top of the main conversation.
- Compaction.
/compacthas to read the conversation it summarizes, so compacting a big session is itself a big request.
How to hit the limit less often#
1. Read the breakdown in /usage#
On a subscription plan, /usage shows what your recent usage went to: skills, subagents, plugins and each MCP server. Under it are recommendations, one per row, such as trimming a named CLAUDE.md or lowering the effort level. Press d or w to switch between the last day and the last week. This is the fastest way to find your own biggest cost.
2. Clear between tasks#
Use /clear when you move to unrelated work. Old context is paid for again on every message that follows. Run /rename first if you want to return to the session with /resume.
3. Shrink what loads every turn#
Your CLAUDE.md files and everything they import load at the start of every session and stay in context for every turn. Anthropic suggests keeping CLAUDE.md under 200 lines and moving instructions for specific workflows into skills, which load only when used.
4. Match the model and effort to the job#
Routine edits run fine on a smaller model. Switch with /model, and lower thinking effort with /effort for simple work. Thinking is billed as output, the most expensive kind of token.
5. Turn off what you are not using#
Run /context to see what fills the window and /mcp to disable servers you are not using. Stop loops and scheduled tasks you no longer need.
The longer list, with the reasons behind each item, is in How to reduce Claude Code token usage. If you use Codex as well, see Codex usage limits.
Doing the memory part automatically#
Step 3 pays on every turn, and it is the one people put off, because splitting a memory file by hand takes time and has to be redone as the file grows. MinniMemoryMCP does the split for you. It sorts your CLAUDE.md or AGENTS.md, MEMORY.md and the files they import into a small part that loads on every turn and files Claude opens only when a task needs them. Every line is kept, and undo puts the originals back.
$ npm i -g minnimemory
$ minnimemory installWith MinniMemoryMCP 1.6.0 on our test memory, the agent went looking in its memory 17 times instead of 89 over the same 87 answers, and the run used 47.2% fewer tokens. Measured 15 November 2026: at API rates, a Claude Code session on our test memory cost $0.47 before MinniMemoryMCP 1.6.0 and $0.29 after on average, 37.7% less (each of the three sessions 33.9% to 40.0% less). A blind judge found 80 of 87 answers correct before and 81 after.
The run counts cost and tokens. Fewer tokens on each turn means each turn draws less from your window; how much longer your window lasts depends on your own work. The method is on the Evidence page.