Read the exact message. If it names a session, weekly, or model limit with a reset time, it is your plan allowance and you wait or switch. If it is a 429 with retry-after, it is throughput and clears in seconds. If it is a 529 Overloaded error, it is Anthropic at capacity, it does not count against your quota, and switching model often works because capacity is tracked per model. Check status.claude.com before anything else.
- A 529 is not your limit and does not count against your quota.
- Capacity is tracked per model, so
/modeloften gets you working during a 529. - The authority is status.claude.com, not a crowd-report site.
- Commit before you do anything else. Being stopped mid-edit is the expensive part.
- A
429retries itself. If it persists, check the credential, not the plan.
Read the exact string first
The messages Claude Code prints, as of August 2026, ordered by how often people hit them.
| Message | Mechanism | Wait | Do |
|---|---|---|---|
You've hit your session limit · resets 3:45pm | Your 5-hour window | Hours | Wait, or add usage credits |
You've hit your weekly limit · resets Mon 12:00am | Your weekly budget | Days | Change how you work, or add capacity |
You've hit your Opus limit · resets 3:45pm | One model family | None | /model and carry on |
API Error: Request rejected (429) | Throughput on your key | Seconds | It retries. Check the credential if it persists |
API Error: Repeated 529 Overloaded errors | Anthropic capacity | Minutes | Check status, or switch model |
If it is Anthropic: 529 and overloaded
The full message reads API Error: Repeated 529 Overloaded errors. The API is at capacity, followed by a note that this is usually temporary and a pointer to the status page. Claude Code has already retried several times before it shows you that, so seeing it means the retries did not help.
- It is not your usage limit and does not count against your quota. Nothing you consumed is lost.
- Capacity is tracked per model. Running
/modeland picking a different one frequently works while one model is saturated. - A related string is throttling, not capacity:
API Error: Server is temporarily limiting requests (not your usage limit). Anthropic retries that one with backoff automatically. Wait briefly.
Where to actually check
| Source | What it tells you | Trust it for |
|---|---|---|
| status.claude.com | Per-component status for claude.ai, the API, Claude Code, Cowork, and the Console, plus 90-day uptime and incident history | The answer |
| Status page Atom or RSS feed, email, Slack, or webhook | The same, pushed to you | Not having to check |
| Downdetector and similar | A count of user reports | Corroboration only |
| Social posts | Whether other people noticed | Nothing on its own |
# is it reachable at all?
curl -sS -o /dev/null -w "%{http_code}\n" https://api.anthropic.com/v1/messages
# 401 means reachable and unauthenticated: the service is up, the problem is you
# 000 or a timeout means network or DNS on your side
# 5xx means it is not youIf it is yours: the next five minutes
Commit whatever is in the tree
git add -A && git commit -m "wip: agent progress before rate limit"An interrupted agent leaves a partial change. Committing it means you can reason about it later instead of reconstructing what happened from memory.
Confirm which credential is live
/statusA stray ANTHROPIC_API_KEY in your environment overrides a subscription login and routes you through whatever tier that key belongs to. It is the single most common reason a 429 keeps recurring on an account that should not be near a limit.
Look at the actual numbers
/usage # plan bars, plus what has been consuming them
# d for the last 24 hours, w for the last 7 daysThe attribution breakdown splits recent usage across skills, subagents, plugins, and individual MCP servers. If one line is a third of your allowance, you have found the problem rather than the symptom.
Pick the escape that matches the cap
/model # only helps a model-scoped limit
/usage-credits # Pro and Max: keep going, billed at standard API ratesUsage credits are the honest answer when the work cannot wait. They bill at standard API rates rather than a flat fee, and they shorten the prompt cache lifetime from about an hour to five minutes, so a long session gets more expensive per turn once you are on them.
Stopping it happening mid-task
Ranked by effect on a subscription.
| Change | Why it works |
|---|---|
/clear between unrelated tasks | Largest single lever. The whole thread is re-sent every turn |
| Stop defaulting to the biggest model | Several times the draw for identical mechanical work |
| Lower the effort level for simple work | Reasoning tokens are billed as output |
| Name files in the prompt | Fewer speculative reads, each of which lands in context |
| Scope test commands | Command output is next turn's input |
| Prune unused MCP servers | Standing overhead on every session |
| Fewer parallel sessions, or a second account | One account is one pool, however many agents draw on it |
Questions people ask
It depends on the exact string. A named session, weekly, or model limit with a reset time is your plan allowance. A 429 is throughput on your API key and clears in seconds. A 529 Overloaded error is Anthropic at capacity and is not your limit at all.
Check status.claude.com, which reports claude.ai, the API, Claude Code, Cowork, and the Console separately and keeps 90 days of incident history. It offers Atom and RSS feeds plus email, Slack, and webhook alerts, so you can be told rather than having to look.
The API is temporarily at capacity across all users. Claude Code retries several times before showing it. It does not count against your usage quota, and because capacity is tracked per model, switching model with /model often gets you working again.
Only as corroboration. It counts user reports rather than measuring the service, so it lags a real incident and conflates outages with client bugs and behaviour changes. Treat status.claude.com as the verdict and a report spike as a second opinion.
Most often a stray ANTHROPIC_API_KEY routing you through a lower-tier organisation than you expect. Run /status to confirm the active credential. If the key is correct, reduce concurrency, run fewer parallel subagents, or request a higher usage tier.
If it named a specific model, yes: switch with /model. Otherwise, usage credits let Pro and Max subscribers continue at standard API rates, and an API key has no session or weekly windows at all. Waiting is the only free option.
No. A 529 means the request never ran and Anthropic states it does not count against your usage limit. The same applies to the temporary throttling message, which is retried with backoff automatically.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Claude Code error reference exact 429, 529, and usage limit strings
- Claude status per-component status and incident feeds
- Manage usage credits for paid Claude plans continuing past the limit at API rates