Claude 529 vs 429 vs Weekly Limit
Three errors, three causes, three fixes
They all feel like "I'm out of quota," but 529, 429, and the weekly limit have completely different causes — confuse them and you'll burn hours retrying the wrong way, or even get flagged for retrying too aggressively.
Three errors, three recovery paths
Overloaded (upstream capacity)
Anthropic's servers are under overall load — unrelated to your request. Safe to retry with exponential backoff.
rate_limit_error (you're throttled)
Your own request rate or concurrency exceeded your account/org limit. Slow down or space out requests and it clears.
"You've hit your weekly limit"
Not throttling — your account's weekly quota is exhausted. Retrying does nothing; you wait for the reset.
Recovery times are wildly different
529 usually clears in seconds to minutes, 429 in tens of seconds of backoff, but a weekly-limit reset can take up to 7 days.
What each error is actually telling you
529 (Overloaded) comes from the server side — Anthropic's API returns it when overall load is high, and the error body's type field reads overloaded_error. It has nothing to do with your request content, token count, or model choice; retrying usually works, but use exponential backoff instead of hammering it immediately, or you add to the very congestion causing the error. 429's error body type is rate_limit_error — meaning your own call rate, concurrency, or tokens-per-minute exceeded your account or organization's limit. This one you can act on directly: lower concurrency, space out polling, or request a higher limit. The weekly limit is a completely different thing: it isn't an HTTP-level throttle at all, it's an account-level quota tied to your Claude Code / subscription plan. The error text literally contains "You've hit your weekly limit" — meaning your allotment for this entire billing cycle is spent. No amount of retrying helps; you either wait for the official reset or buy/activate extra capacity.
Why this distinction matters more right now
On 2026-08-18 Anthropic announced the +50% weekly quota boost is extended through 2026-08-31 — many accounts have had noticeably more headroom than usual during this window. That also means once the boost lapses after Aug 31, accounts that rarely hit the weekly limit before may run into it for the first time. Confusing it with 529/429 usually means hammering retries against an account-level quota problem — which not only doesn't help, but overly aggressive retrying can get flagged as abnormal traffic.
Timeline
529/429 are standard Anthropic API status codes that have existed as long as the API; public docs and developer discussion consistently describe what they mean.
Anthropic officially announces the +50% weekly quota boost is extended through 2026-08-31, temporarily masking the weekly limit some accounts would otherwise have hit.
The temporary boost lapses; whether it continues hasn't been announced. Accounts that rarely hit the weekly limit before should watch for this error again.
Confirmed vs. common misreadings
Confirmed
The error body's type field (overloaded_error / rate_limit_error) and the literal Claude Code CLI text "You've hit your weekly limit" show up consistently across public developer discussion and Anthropic's own documented status descriptions — the most reliable signals for telling these apart.
Common misreading
A claim that circulates online is that retrying enough on a 529 will silently get you routed to a smaller or quantized model. We found no official statement or credible first-hand evidence for this. 529 only signals capacity pressure — it doesn't mean your request got rerouted to a different model, so treating retry count as a way to "switch models" isn't reliable.
What to actually do: three very different paths
529 / 429: your client can fix it
Retry 529 with exponential backoff (1s → 2s → 4s… plus jitter). For 429, do the same but also check for a retry-after header and consider lowering concurrency or requesting a higher limit. Neither needs human support — most SDKs' default retry logic already covers both.
Weekly limit: your client can't fix it
When the error text says "You've hit your weekly limit," the only useful move is checking the reset time shown on your account, or temporarily switching to a model/provider whose quota isn't exhausted. Routing the same key through QCode to a different vendor's model lets you keep shipping without waiting for the reset.
A 3-step way to tell them apart
Step 1: check the HTTP status — 529, 429, or no standard status code at all, just plain text. Step 2: check the error body's type field (for the two HTTP-level errors) or the literal error text (the weekly limit is account-level, and usually isn't a standard HTTP error body at all — it's a message surfaced by the CLI or dashboard). Step 3: check your dashboard — if it shows this period's quota at zero with a reset time days away, that's the weekly limit, not a transient server blip.
What to do on QCode
Whichever one you hit, one QCode key can route requests to model families outside Claude — GPT, Gemini, GLM, Kimi, DeepSeek, Qwen and more — as a stopgap, so you keep shipping without waiting for 529 to clear or the weekly limit to reset.
FAQ
Are 529 and 503 the same thing?
Not quite. 529 is Anthropic's API-specific "overloaded" status; 503 more commonly shows up at a generic gateway/load-balancer layer as "service unavailable." Both are worth retrying, but 529 specifically points to the model service's own capacity.
429 and the weekly limit are both 'limits' — what's the real difference?
429 is a rate/concurrency limit — a momentary throttle within a sliding window that usually clears after tens of seconds to a few minutes of backoff. The weekly limit is a total quota for an entire billing cycle under your subscription; once it's spent you wait for the official reset time, and slowing down your request rate won't bring it back early.
Should I retry a 529 with a smaller model?
No need. 529 has nothing to do with model size or parameters — it's an overall server capacity issue. Switching models might just happen to hit a less-loaded capacity pool, which isn't a reliable fix. Exponential backoff is what actually works.
How do I tell whether I hit 429 or the weekly limit?
Check the error text and body. 429 usually comes with a standard HTTP error body and type: rate_limit_error. The weekly limit usually shows up as a plain message, literally "You've hit your weekly limit", and your dashboard will show this period's quota exhausted with a reset time days — not minutes — away.
How long should backoff waits be?
A common pattern is starting at 1 second and doubling after each failure (1s, 2s, 4s, 8s…) with a bit of random jitter so multiple retries don't pile up into a new spike. Most official SDKs already build this in — copy that pattern if you're rolling your own.
Is there anything I can do ahead of hitting the weekly limit?
Watch the remaining-quota trend on your account dashboard so you can anticipate when it'll run out. As a stopgap, route non-critical work through QCode to other model families and save your remaining Claude quota for what actually needs it.
Sources
The status codes and error-body fields (type: overloaded_error / rate_limit_error) come from Anthropic's public API documentation on status handling; the literal "You've hit your weekly limit" error text comes from screenshots and discussion Claude Code users have shared publicly; the +50% weekly boost extension through 2026-08-31 comes from Anthropic's official account announcement on 2026-08-18. This page was checked against these sources on 2026-08-27.
Don't let one account block your delivery
One QCode key routes to GPT, Gemini, GLM, Kimi, DeepSeek and more — no waiting for 529 to clear or the weekly limit to reset.
Related reading
What 529 Overloaded actually means
A deeper standalone look at 529, including how it differs from 503.
What to do about 'Claude usage limit reached'
How to tell the 5-hour rolling window from the weekly limit, and what to do about each.
Codex has ghost quotas too: status green, request still rejected
The same kind of problem shows up on Codex — how to diagnose it there.
This page is technical explanation, not an official Anthropic statement. Specific quota rules and reset times are governed by your account dashboard and Anthropic's official documentation.