API limits
Anthropic rate limits: API 429s, tiers, and how they differ from Claude usage caps
Last reviewed July 2026.
Short answer: Anthropic rate limits are API-side ceilings — how many requests and tokens you can send per minute (and related windows) on your API tier. Hitting them usually surfaces as HTTP 429. They are not the same thing as Claude Code’s subscription 5-hour window and weekly cap. If you are capped inside the Claude Code CLI on a Pro/Max plan, you want Claude Code usage limits. If your backend or script is getting 429s from api.anthropic.com, you are in the right place.
Two different limit systems (read this first)
| Claude subscription usage limits | Anthropic API rate limits | |
|---|---|---|
| Where you see them | Claude / Claude Code product on a Pro or Max plan | API keys, SDK calls, server-side integrations |
| What is metered | Usage over a rolling 5-hour window + weekly ceiling | Request rate, token throughput, tier caps |
| Typical signal | In-product “usage limit” / wait-until messaging | HTTP 429, rate-limit headers, provider errors |
| Primary fix | Wait for the clock, upgrade plan, or cut token burn | Back off, lower concurrency, smaller payloads, higher API tier |
Mixing them up wastes time: upgrading Max will not fix a misconfigured API client hammering requests, and raising an API tier will not refill a spent Claude Code weekly cap.
What “rate limit” means on the API
Anthropic’s API enforces throughput limits so a single key cannot overwhelm the service. As of July 2026, the public docs describe limits that scale with usage tiers — newer or lower-volume keys generally see tighter requests-per-minute and tokens-per-minute (or day) ceilings; higher tiers unlock more headroom. Exact numbers move; always check Anthropic’s current rate-limit documentation for the figures that apply to your account.
When you exceed a limit, the API responds with 429 Too Many Requests (sometimes accompanied by retry guidance or rate-limit headers, depending on client and moment). That is a signal to slow down, not a signal that your Claude Max weekly bucket is empty.
Common causes of 429s
- Bursting many parallel requests (agent fan-out, batch jobs without a queue)
- Very large prompts that burn the tokens-per-minute budget in a few calls
- Retries without backoff that amplify a temporary limit into a prolonged outage
- Sharing one API key across many environments without coordination
What to do when you get a 429
- Back off and retry with jitter — immediate tight loops make it worse.
- Lower concurrency — serialize or cap parallel Claude calls.
- Shrink payloads — send only the context the model needs.
- Confirm your API tier — as of July 2026, higher tiers carry higher throughput; verify in Anthropic’s console/docs rather than assuming.
- Confirm you are not on the wrong mental model — if the failure is inside Claude Code on a subscription with a wait timer for a usage window, switch to the usage limits hub or the 5-hour limit guide.
Subscription usage limits (the other intent)
Claude Code on Pro/Max is gated by usage limits: a rolling 5-hour window and a weekly ceiling, with larger room on Claude Max. Those caps are about how much agent usage your subscription includes over time. They are the pain Hypermile focuses on — forecasting when you will hit the wall and trimming noisy prompt content so the plan lasts longer.
If your question is “when does my Claude Code cap reset?” start at Claude Code usage limits, not here.
Where Hypermile fits (and does not)
Hypermile is a cross-platform usage copilot for Claude Code and Codex subscription workflows: local audit, forecast, and reversible compression. It is not an Anthropic API gateway and it does not raise your API rate-limit tier. If your problem is 429s from your own API integration, fix client backoff and tier first. If your problem is burning a Claude Code weekly cap on Windows/WSL, measure with the free audit and consider the app.
Next steps
Still on the subscription side? Read Claude Code usage limits. Debugging agent burn on Windows? See the Claude Code on Windows hub, then download Hypermile if you want a forecast and compression layer for local Claude Code/Codex usage — not for API tier increases.
Frequently asked questions
What are Anthropic rate limits?
They are caps on how fast and how much you can call the Anthropic API — typically requests per minute, tokens per minute or day, and related tier ceilings. Hitting them returns errors such as HTTP 429. They apply to API keys and usage-based API billing, not to the Claude consumer subscription's 5-hour/weekly windows.
What does HTTP 429 mean from Anthropic?
429 means you hit a rate limit: too many requests or too many tokens in the current rate window for your API tier. Back off, retry with jitter, and reduce concurrency or payload size. It is not the same message as Claude Code's subscription 'usage limit' screens.
Are Anthropic rate limits the same as Claude Code usage limits?
No. Subscription Claude Code limits are a rolling 5-hour window plus a weekly cap on your Claude plan. Anthropic API rate limits are request/token throughput ceilings on an API key tier. Different products, different meters, different fixes.
How do I raise Anthropic API rate limits?
As of July 2026, Anthropic ties higher throughput to API usage tiers and account status. Sustained successful usage and moving up tiers generally raises limits; exact thresholds are in Anthropic's rate-limit documentation and may change. Reducing peak concurrency also avoids 429s without a tier change.