Fix a Claude rate limit during summarization.

Anthropic's Claude rate limits are enforced across four dimensions: requests per minute (RPM), input tokens per minute (ITPM), output tokens per minute (OTPM), and a 5-hour rolling window on some workloads. The limits depend on your Anthropic tier (1-4) and the specific Claude model — Sonnet and Opus have lower throughput ceilings than Haiku. The fastest fix is to verify the exact limit you hit, wait for the window to reset (60 seconds for per-minute, longer for tier-level), then either reduce source size, request a tier upgrade, or switch providers in Reduz while Claude's quota catches up.

Check these first

• You've hit the requests-per-minute (RPM) limit — common after rapid retries on a low tier.
• A long PDF or transcript exceeded the input-tokens-per-minute (ITPM) cap — Sonnet has notably lower ITPM than Haiku.
• Your generated output is hitting OTPM during a long summary or full-notes generation.
• Your Anthropic tier is too low for your workload — Tier 1 is the most restrictive.
• A prompt cache miss made the request consume more billable tokens than expected (caches reset after 5 minutes of inactivity).

Fix it in this order

1. 1

   Read the exact limit name in the error response

   Anthropic returns specific error messages: "rate_limit_error" with body text identifying RPM, ITPM, OTPM, or organization-level. The fix is different for each — read the error before retrying blindly.

2. 2

   Wait at least 60 seconds for per-minute limits

   RPM, ITPM, and OTPM all reset on a rolling 60-second window. Wait that long before retrying. Aggressive retries during the window keep the limit pinned.

3. 3

   Check your usage at console.anthropic.com

   Open console.anthropic.com/settings/limits. Verify your tier, current usage, and the per-model RPM/ITPM/OTPM caps. If you're consistently at the ceiling, request a tier upgrade or scale workload.

4. 4

   Switch to a Haiku-tier Claude model

   Claude Haiku 4.5 has substantially higher throughput than Sonnet 4.6 or Opus 4.7. For high-volume daily summaries, Haiku is often the right default. In Reduz settings under Anthropic, switch the active model and retry.

5. 5

   Reduce source size for big PDFs and transcripts

   Long sources push ITPM to the ceiling in a single request. Use selected-text mode for one section, summarize a PDF page range instead of the whole document, or break a 90-minute video summary into shorter passes.

6. 6

   Request a tier upgrade for sustained workload

   Anthropic auto-upgrades tiers based on consistent usage and spend. For production-grade workloads, submit a manual tier-increase request through the console. New tiers unlock substantially higher RPM/ITPM/OTPM ceilings.

7. 7

   Switch to another provider while you wait

   If urgent work is blocked, switch Reduz to OpenAI, Google Gemini, DeepSeek, or xAI Grok. The same source runs through a different provider with different limits. Switch back to Claude once your window resets.

Diagnosis

Anthropic-controlled

Rate limits are set by Anthropic for your tier, your account, and the specific model. Reduz cannot raise them — it can only reduce request size or route to a different provider.

Model-specific

Claude Sonnet 4.6 and Opus 4.7 have lower throughput than Haiku 4.5. A workload that hits limits on Sonnet may run fine on Haiku.

Token-driven for long sources

A 30-page PDF can consume 30,000+ input tokens. On Tier 1 ITPM (40k for Sonnet), one large PDF can saturate the per-minute window immediately.

Tier-dependent

Tier 1 (small spend) has restrictive limits. Tier 3-4 unlocks RPM/ITPM/OTPM ceilings that fit production summarization workloads. Auto-upgrade happens with consistent usage; manual requests are also accepted.

Prompt cache helps if reused

Anthropic offers prompt caching that reduces input-token cost on cache hits. For repeat summaries of the same source with different prompts, enabling caching reduces effective ITPM usage substantially.

Keep Claude for quality, switch when blocked

Claude's summarization quality is excellent — especially Sonnet 4.6 for long technical content and structured outputs. The trade is that on lower tiers, rate limits can interrupt high-volume work. Reduz can't raise your Anthropic limit — that's on Anthropic — but it lets you keep your workflow moving while the limit resets. Switch to OpenAI, Google Gemini, DeepSeek, or xAI Grok for the next few summaries; switch back to Claude once you're past the window. For sustained high-volume use, request a tier upgrade through the Anthropic console — Tier 3-4 limits handle production summarization workloads comfortably. Hosted Free in Reduz also gives you a Claude-independent fallback that doesn't touch your Anthropic account at all.

Frequently asked questions