Omniscio

OmniscioGuidesClaude Code overloaded errors

Claude Code overloaded and 429 errors explained

Two messages stop Claude Code when nothing is wrong with your code: a 529 Overloaded error, which means Anthropic's API is at capacity, and a 429 Request rejected error, which means a rate limit is holding your requests back. This page covers those two, the retries Claude Code makes before it ever shows you one, and the steps Anthropic documents for each. Every message and every step below is sourced and dated in the list at the end.

Last checked against Anthropic's documentation on October 23, 2026. By the Omniscio team. Updated .

Which message did you see?

Claude Code prints one of these when Anthropic's API is at capacity, or when a rate limit is holding your requests back. Each has its own section further down.

API Error: Repeated 529 Overloaded errors. The API is at capacity — this is usually temporary. Try again in a moment. If it persists, check https://status.claude.com.

The 529. The API is at capacity for everyone, and Claude Code has already retried several times before showing this.

Opus is experiencing high load, please use /model to switch to Sonnet

The same overload, said in terms of one model, because capacity is tracked per model.

API Error: Request rejected (429) · this may be a temporary capacity issue. If it persists, check https://status.claude.com.

The 429. A rate limit is configured somewhere on the way to the API, and it is your key or your provider project rather than your plan.

Looking for something else? The Claude Code error messages guide covers the rest, from a prompt that is too long to a login that has expired.

Our servers are currently overloaded. Please try again later.

If that is the sentence on your screen, Anthropic's Claude Code documentation does not list it. On October 23, 2026 we searched that wording word for word through Anthropic's Claude Code error reference, its API error list, its API rate limits reference and its Claude apps gateway configuration, and it appears in none of them. A phrase search of Anthropic's help centre turns up only unrelated articles.

Anthropic's documentation does not say which product shows the sentence. The situation it describes is the one the 529 covers, and the steps for that are the same either way: check the status page, send the message again after a few minutes, and switch models with /model if you need to keep working now.

Sources: the four Anthropic pages named above and a phrase search of Anthropic's help centre, all read October 23, 2026. The status page is status.claude.com.

Working out which limit stopped you

Several different things stop Claude Code from getting an answer, and they need different responses. Anthropic documents each one separately, so the message on your screen is the quickest way to tell them apart.

  • A 529 Overloaded error means the API is at capacity across all users. Anthropic's error reference says a 529 is not your usage limit and does not count against your quota.
  • A 429 Request rejected error means you hit the rate limit configured for your API key, your Amazon Bedrock project or your Google Cloud project.
  • A plan usage limit is a third message again. It belongs to a Claude subscription, it names a reset time, and the error messages guide has it.
  • A monthly spend limit names whose limit blocked you and where to raise it, and it is on that same guide.
  • A 500 is a failure inside the API rather than a capacity limit. Anthropic's own wording for it says it is not caused by your prompt, settings or account.
API Error: 500 Internal server error. This is a server-side issue, usually temporary — try again in a moment. If it persists, check https://status.claude.com.

The single thing worth checking first, whichever of these you have, is which credential the request went out with. Run /status. Anthropic notes that an approved ANTHROPIC_API_KEY in your environment takes precedence over your subscription login, which is how a 429 appears for someone who believed they were on a plan.

Sources: Anthropic's Claude Code error reference and Anthropic's API error list, read October 23, 2026.

What Claude Code tried before it showed you the message

The message is the last thing that happens, not the first. Anthropic's error reference says Claude Code retries transient failures up to 10 times with exponential backoff before showing an error, and that by the time you see one it has already made whatever retries apply to that failure.

While it retries, the spinner carries a countdown:

Retrying in Ns · attempt x/y

The label in front of the countdown names the reason from the first attempt for failures you can act on straight away, such as a network that is down or a rate limit. For other failures it reads API error at first and switches to the specific reason from the third attempt. Once the reason is shown, a 529 adds the address to check beneath it: status.claude.com on the Anthropic API, or the provider or gateway host named in the message on another configuration.

A stalled response is handled differently again. After 20 seconds with no data arriving on the stream, the spinner shows this, and the request has not failed yet:

Waiting for API response · will retry in … · check your network

Three cases are deliberately not retried, and each one explains an error that a reader might expect to have cleared itself. A failure that arrives after Claude has already started answering is not re-sent, because re-sending could run the same tool calls twice; Claude Code keeps what was finished and carries on from there. A certificate validation failure is reported on the first attempt so the certificate setup can be fixed straight away rather than after several minutes of retrying. And a failure that arrives after the whole response is complete needs no retry at all, so the turn ends normally.

Sources: the automatic retries, "what you see while Claude Code retries or waits", and tune retry behavior sections of Anthropic's Claude Code error reference, read October 23, 2026.

The 529 Overloaded message in depth

A 529 is an overloaded_error from Anthropic's API: the service is temporarily overloaded, and the error list adds that it can appear when the API sees high traffic across all users. Two consequences follow, and Anthropic documents both.

  • It is not your allowance. The error reference says a 529 is not your usage limit and does not count against your quota, so the minutes it costs you are not minutes taken out of your plan.
  • Capacity is tracked per model. That is why changing models is the step Anthropic gives for keeping work moving, and why Claude Code prompts you to do it when one model is under particularly high load. On Fable models that prompt names Fable instead of Opus.

The trailing sentence of the message is the part that changes with how you connect. On a Bedrock, Google Cloud's Agent Platform or Microsoft Foundry configuration it names that provider's status page, and with a custom base URL it names the gateway host.

In a session the Claude Desktop app runs, such as the Code tab or Cowork, the same overload reads in terms of the model only, and the app's model picker is where you switch.

Sources: the 529 entry and the retry sections of Anthropic's Claude Code error reference, and the status-code table in Anthropic's API error list, read October 23, 2026.

The 429 message in depth

A 429 is a rate limit, and the error reference names the three things it can belong to: your API key, your Amazon Bedrock project, or your Google Cloud project. If you signed in with a Claude subscription and never set a key, run /status before anything else, because an approved ANTHROPIC_API_KEY in your environment routes requests through that key instead of your login, so a 429 can arrive for a key you did not mean to use.

Rate limits on an Anthropic API key are counted per model class in three units: requests per minute, input tokens per minute, and output tokens per minute. Limits are set at the organization level and defined by a usage tier that rises as an account builds history, and the Claude Console shows both the tier and the limits in force on its rate limits page, which is also where a higher tier is requested.

One 429 is not a throttle at all. The rate limits reference says a request also comes back 429 once an organization reaches its monthly spend cap, and that response carries no retry-after header, so retrying fails until access resumes.

When a 429 does carry retry-after, the header holds the number of seconds to wait and earlier retries fail, so waiting exactly that long is the documented move. Where Claude Code can act on a 429 by itself it already has, because temporary throttles are in its retry list. A gateway's spend-limit 429 is the exception: the reference says it is not a throttle, so it is not retried.

If the text after the dot looks like Too Many Requests rather than Claude Code's own wording, a proxy, load balancer or gateway between you and the API answered with a page of its own, and Claude Code shows that page's title. Before version 2.1.281 it printed the page's raw markup there instead, which is worth knowing if you are reading an older copy of the message.

A sharp rise in usage can also produce 429 errors from acceleration limits, and the remedy Anthropic gives is to increase traffic gradually rather than in a step. A Claude Code workspace has its own limits too, checked separately from the rest, and requests over them can return a 429 with a retry-after header.

If you pay per token rather than by subscription, the cost calculator turns a session's token counts into a figure.

Sources: the 429 entry in Anthropic's Claude Code error reference, and Anthropic's API error list with Anthropic's rate limits reference for the tiers, the counts, the headers and the spend cap, read October 23, 2026.

Fast mode's separate rate limit

If you have fast mode switched on, its rate limit is separate from the standard one. Anthropic's fast mode page says every supported Opus model shares a single fast mode rate limit pool, so usage on any of them draws on the same limits.

Reaching that limit is documented as a sequence rather than an error you have to clear: fast mode falls back to standard speed on its own, the ↯ icon turns gray to show the cooldown, you carry on working at standard speed and pricing, and fast mode re-enables itself when the cooldown expires. Running /fast turns it off by hand if you would rather not wait for the cooldown.

There is one case where it is switched off for you. If your usage credits run out mid-session, Claude Code retries each rejected fast mode request at standard speed and pricing, so work continues and there is no cooldown, and an interactive session shows this notice and leaves fast mode off for the rest of the session:

Fast mode disabled · usage credits exhausted

Your saved preference does not change, so /fast is what turns it back on. Fast mode is also a research preview, and Anthropic notes that its availability and its pricing can change.

Source: Anthropic's fast mode page, read October 23, 2026.

What actually changes the outcome

Every step below is one Anthropic's own pages give for these two messages.

  • For a 529, check status.claude.com first. The message names where to look, and an incident posted there is the whole answer.
  • For a 529, send the message again after a few minutes. The error reference gives this as the step, and your original message is still in the conversation, so a long prompt does not need pasting again.
  • For a 529, run /model and switch to another model to keep working now, because capacity is tracked per model.
  • For a 429, run /status and confirm the active credential is the one you meant to use.
  • For a 429, read your provider console for the limits in force and ask for a higher one if you need it.
  • For a 429, reduce what you send at once: fewer parallel subagents, or a smaller model with /model for high-volume scripted runs.
  • For a 500 that keeps happening with no incident posted, run /feedback so Anthropic has your request details to work from.
  • In an unattended session such as a CI job, CLAUDE_CODE_RETRY_WATCHDOG=1 retries 429 and 529 capacity errors without the usual ceiling instead of stopping after the retry count runs out.
  • Also in unattended sessions, CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS caps how long one request spends waiting out a 429 or a 529, and CLAUDE_CODE_RETRY_WATCHDOG_USAGE_LIMIT_RECHECK_MS sets how often a waiting request checks whether a usage limit has already lifted.
  • To change how far apart the retries of a 529 sit, CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS sets the starting gap and CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS caps it.
  • CLAUDE_CODE_MAX_RETRIES changes the number of attempts, which is 10 by default; lowering it surfaces failures sooner in a script, but it does not change anything on the API's side.
  • If every request fails with a 529 or a 500 and the base URL points at an LLM gateway, ENABLE_TOOL_SEARCH=false is the check Anthropic gives for a gateway that rejects the tool search fields Claude Code sends by default.

What none of those lists offers is a way to make capacity appear, and four things that get tried instead are worth ruling out. Restarting does not clear a 529, because the retries were already spent before the message appeared and the capacity in question belongs to the API across all users rather than to your session. Waiting for your plan's reset time does not clear one either, because the same error reference says a 529 is not your usage limit. Neither the 529 nor the 429 step list offers /compact, /clear or a smaller prompt as a way out; those belong to the context errors, which the error messages guide covers. And neither list asks for a reinstall: claude update appears in other entries of the reference, where a version note names the build that changed, and it is not part of either step list here.

Sources: the 529 and 429 step lists, the automatic retries and tune retry behavior tables in Anthropic's Claude Code error reference, and the "Reaching your spend cap" section of Anthropic's rate limits reference, read October 23, 2026.

Running several sessions at once

When you run a group of sessions, a subscription login that has used up its allowance bites hardest, because one account at its usage limit can strand a whole row of them together. Omniscio, the desktop app we make for running many Claude Code sessions at once, spreads new sessions across the login accounts by how much each has left, and when one account reaches its usage limit only that account's sessions are affected: each one separately re-picks a login that still has room instead of everything waiting on the same reset. A session whose pool is full is shown as waiting rather than live, and the status list and the account pool's ordering are both documented in full.

Two guides cover the surrounding work: running multiple Claude Code sessions and running many coding agents at once. The features page lists the rest, and the download page has the installers.

Sources

Every message, every meaning and every step above rests on one of these, and the date beside each is the day we read it.

Anthropic's error reference moves with nearly every Claude Code release and keeps notes such as "Before v2.1.281" and "Before v2.1.199". Read it again before you trust a copy of this page from a few months back.

Common questions

Is "our servers are currently overloaded. Please try again later." a Claude Code message?

Not one Anthropic documents. That exact wording is not in Anthropic's Claude Code error reference, its API error list, its rate limits reference or its Claude apps gateway configuration, and a phrase search of Anthropic's help centre turns up only unrelated articles. The closest message Claude Code does print is the 529 Overloaded line, which says the API is at capacity for everyone and names the status page to check.

Does a 529 Overloaded error count against my usage limit?

No. Anthropic's error reference says a 529 is not your usage limit and does not count against your quota. It means the API is at capacity across all users, so the time it costs you is not spent out of your plan.

Is a 429 the same as hitting my Claude plan's limit?

No. A plan's usage limit is a subscription allowance that lifts at the reset time the message names, and Anthropic keeps it in a separate entry. A 429 is the rate limit configured for your API key, your Amazon Bedrock project or your Google Cloud project, and the remedy sits in that provider's console rather than in your plan.

Why does the error appear only after Claude Code has been quiet for a while?

Because it retries first. Anthropic's error reference says Claude Code retries transient failures up to 10 times with exponential backoff before showing you an error, and the spinner counts that down as Retrying in Ns · attempt x/y.

Does switching models clear a 529 Overloaded error?

Yes, and it is the step Anthropic gives for keeping work moving, because capacity is tracked per model. Switching does not help with a limit that covers every model, so it is worth checking which message you have before you switch.