Read the message before doing anything else. "You have hit your session limit" is the rolling 5-hour window and clears on its own. "You have hit your weekly limit" needs a change, not patience. "You have hit your Opus limit" is model-specific and switching model with /model keeps you working immediately. A 429 is not a plan limit at all: it is API throughput on your key or cloud project. Run /usage to see the bars and reset times, /usage-credits to buy headroom, and commit whatever is in your working tree first.
- The message names the limit: session, weekly, Opus, or a 429. Four problems.
- A weekly or session limit is shared across models, so
/modeldoes not restore access. - An Opus limit is the exception:
/modelgets you working again in one keystroke. /usageshows the bars and reset times./usage-creditsbuys headroom past them.- Commit first. Being cut off mid-edit is the expensive part.
- Usage is shared with Claude chat and Cowork on the same account.
Which limit did you hit
Claude Code tells you exactly which ceiling you reached and when it resets. The wording matters more than it looks, because three of these four are unrelated mechanisms.
Message wording as of August 2026, and what each one means.
| You see | It is | Wait? | Do |
|---|---|---|---|
You've hit your session limit · resets 3:45pm | The rolling 5-hour window | Hours | Take the break. Review the diff. |
You've hit your weekly limit · resets Mon 12:00am | The weekly plan allowance | Days | /usage-credits, or change how you work |
You've hit your Opus limit · resets 3:45pm | A model-specific allowance | No | /model to Sonnet and keep going |
API Error: Request rejected (429) | API throughput on your key or cloud project | Seconds | Lower concurrency; check the provider console |
API Error: Repeated 529 Overloaded errors | Service capacity, not you | Minutes | Check the status page; try another model |
The 5-hour window is rolling, which people misread constantly. Capacity returns continuously as older usage ages past the five-hour mark rather than all at once at a round hour. The message gives you the point at which you are back in business, not the start of a fresh allowance.
The next five minutes
Commit whatever is in the tree
git add -A && git commit -m "wip: agent progress before usage limit"An interrupted agent leaves a partial change. Committing it means you can reason about it later instead of reconstructing what happened from memory.
Look at the actual numbers
/usageYou get plan usage bars with reset times, plus a breakdown attributing recent usage to skills, subagents, plugins, and individual MCP servers, and flags for behaviours such as long context or cache misses that account for 10 percent or more of recent usage. Press d or w to switch between the last 24 hours and the last 7 days.
If it names Opus, switch model
/model sonnetThe Opus allowance is separate. One keystroke and you are working again, on a model that handles most coding work well and consumes the shared allowance far more slowly.
If it is session or weekly, decide between waiting and paying
/usage-creditsUsage credits let you continue past the plan allowance at standard API rates. On Pro and Max it opens your billing settings. On Team and Enterprise with billing access it opens organization usage settings; without billing access it sends a request to your admins after you confirm.
Why usage climbs faster than your activity
People hit these limits after what feels like an hour of light work, then conclude the allowance shrank. It did not. A long-running session draws usage in ways that are invisible from the transcript.
- Every turn re-sends the whole conversation. A one-line question in a session that has been open all day still carries the entire history. Prompt caching makes that cheap, not free.
- Cache misses after a break reprocess everything. The cache lifetime is an hour on a subscription and drops to five minutes once you are drawing on usage credits, or on an API key. Lunch costs you a full re-read.
- Every tool result is another request. A thirty-step task is thirty requests, each carrying the growing prefix.
- Scheduled tasks fire while the session is idle, sending your full context each time.
- Agent teammates keep consuming until they exit. Agent teams use roughly 7x the tokens of a standard session when teammates run in plan mode.
/compactis itself a large request, because it reads the conversation it summarises./clearcosts nothing.
Stopping it recurring
Ranked by effect on a typical week.
| Change | Effect |
|---|---|
| Default to Sonnet, reserve Opus for hard reasoning | Largest single lever |
/clear between unrelated tasks | Stops paying for stale context on every turn |
Lower the effort level with /effort for simple work | Thinking tokens bill as output |
| Name files in your prompt | Fewer speculative reads before it can act |
| Scope test commands so they print less | Command output is next turn input |
| Delegate verbose reads to a subagent | The log stays in the subagent context |
| Disable MCP servers you no longer use | Standing overhead removed from every session |
| Fewer parallel sessions and teammates | Consumption scales with them, roughly linearly |
| Buy usage credits | Works, and is the answer that costs money |
A 429 is not a usage limit
These two get filed under "rate limit" and neither is a plan allowance. Both resolve without you changing anything about how you work.
| 429 | 529 Overloaded | |
|---|---|---|
| Cause | Throughput cap on your API key or cloud project | Service capacity across all users |
| Scope | Your organization | Everyone on that model |
| Timescale | Seconds to minutes | Minutes |
| First check | /status for the active credential | status.claude.com |
| Lever | Lower concurrency, request a higher tier | /model, since capacity is tracked per model |
# fewer simultaneous tool calls per session
export CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=2
# and check which credential is actually billing
# (inside a session)
/statusQuestions people ask
You exhausted the rolling five-hour usage window on your plan. It is shared with Claude chat and Cowork on the same account, and it clears on its own; the message shows the reset time. Because the window covers every model, switching model does not restore access.
The message states the reset time, and /usage shows the same thing with bars for each window. The five-hour window is rolling, so headroom returns gradually as older usage ages out rather than all at once. The weekly window resets at a fixed time assigned to your account.
Session and weekly windows are shared across all models, so changing model does not give you back allowance you already spent. The one exception is the Opus limit, which is model-specific: after that message, /model to another model keeps you working immediately.
Yes, with usage credits. Run /usage-credits to turn them on or request them from an admin. They bill past the plan allowance at standard API rates. Without them, a subscription blocks until the window reopens.
Usually Opus left as the default, a session left open for hours so every turn re-sends the whole conversation, scheduled tasks firing while you are idle, or agent teammates still running. Run /usage and read the breakdown, which flags any behaviour accounting for 10 percent or more of recent usage.
A 429 is throughput on your API key or cloud project, measured per minute, and it resolves in seconds. A usage limit is a plan allowance measured over five hours or a week. A 529 is neither: it means the service is at capacity, which you can sometimes route around with /model.
A plan change raises the allowance, but the durable fix is usually model choice and session hygiene, which cost nothing. If you are on Opus by default and never clear between tasks, the next tier buys weeks rather than months.
Yes. On Pro, Max, Team, and Enterprise the allowance is shared across Claude Code, Claude chat, and Cowork on the same account, which is why a heavy chat morning can shorten your coding afternoon.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.