Rate limits and usage tiers

API rate limits are organised into three standard tiers:

Tier	Intended for
Start	Entry-level, getting going
Build	Growing applications
Scale	Highest standard limits, production workloads

Beyond Scale, limits are arranged directly with Anthropic’s sales team.

Why this page does not list the numbers

Per-tier, per-model request and token limits change, and a stale table here would be worse than no table — someone would size a launch against it. The current values are in the rate limits documentation and in the Console under Settings → Limits.

That is a deliberate editorial choice, not an omission: see how we source.

What rate limits are not

• Not plan limits. Pro, Max and Team limits govern the Claude apps and are a completely separate system.
• Not spend limits. Hitting a rate limit throttles you; it does not cap your bill. Spend controls are a separate feature, available on Enterprise and through the Console.
• Not shared across providers. Bedrock, Google Cloud and Foundry apply their own quotas.

Reducing pressure rather than raising limits

Batching moves work off the synchronous path entirely. Prompt caching cuts input token consumption, which is what most token limits actually measure. Choosing Haiku 4.5 where it suffices does both.