Rate limits
Claude Fable 5 rate limits
Three per-minute axes, one pool per model, and a set of numbers noticeably smaller than the rest of the Claude lineup.
The 30-second version
Fable 5 is limited on requests per minute, input tokens per minute and output tokens per minute, enforced at the organization level and applied separately per model. On the Start tier that is 1,000 RPM, 500,000 ITPM and 100,000 OTPM; on Build, 2,000, 1,500,000 and 300,000; on Scale, 4,000, 4,000,000 and 800,000. Cache reads do not count toward ITPM, which is the largest lever you control.
How the limits are structured
Four things determine the ceiling you actually hit, and only two of them appear on the pricing page.
Fable 5 has its own pool, and it is smaller
Limits are applied separately for each model, so you can drive several models to their respective limits at once. Fable 5 sits far below the rest: 500,000 ITPM on Start against 2,000,000 for Opus 5, Sonnet 5 and Haiku 4.5. Opus 4.x models share one combined bucket, while Opus 5 and Fable 5 each get their own.
Three independent per-minute axes
RPM, ITPM and OTPM are enforced independently, so exceeding any one of them returns a 429. The limiter uses a token bucket that replenishes continuously rather than resetting at fixed intervals, which is why short bursts can trigger a 429 while your per-minute average sits comfortably under the number.
Usage tiers, not a pay-as-you-go dial
Organizations are placed on the Start, Build, Scale or Custom tier automatically based on usage history and account standing, and move up over time. Start, Build and Scale each carry a monthly spend cap of $500, $1,000 and $200,000. New organizations may begin below the published numbers while account history is established.
Workspace limits sit underneath organization limits
You can set lower spend and rate limits per workspace to stop one team from consuming everything. Organization-wide limits always apply even if workspace limits add up to more, and the anthropic-ratelimit headers report whichever limit is most restrictive at that moment.
Raise your effective throughput in this order
Caching first, traffic shape second, a limit increase request last. The request form asks about your cache rate, so do the caching work before you file it.
- 1
2. Cache everything that repeats
For most models, cache_read_input_tokens do not count toward ITPM. Cache system instructions, tool definitions, large context documents and conversation history. With an 80 percent cache hit rate against a 2,000,000 ITPM limit, you can process 10,000,000 total input tokens per minute.
- 2
1. Find your real numbers
Open the Limits page in the Claude Console to see your tier and current limits, or read them programmatically with the Rate Limits API. Use the Usage page to see hourly peak uncached input tokens per minute and your cache hit rate against your configured limit.
- 3
4. Request an increase with evidence
Use Request rate limit increase on the Limits page. Bring your peak input and output tokens per minute per model and the share of your input that is cached or repeated context. For urgent needs, contact Anthropic support directly.
- 4
3. Smooth the traffic shape
Queue and pace requests instead of fanning out, and ramp new workloads up gradually rather than switching them on at full volume, which can trigger acceleration limits. For bulk work that tolerates latency, move it to the Message Batches API, which has its own separate limits.
Already getting 429s?
The 429 page walks through identifying which of the three axes ran out and what to do about each one.