Anthropic Rate Limit Tiers: How the Usage Tier Ladder Works and What Each Tier Changes (2026)

Anthropic rate limit tiers are the usage levels that decide how much an organization can use the Claude API. According to Anthropic's rate limits documentation, the current ladder is Start, Build, Scale and Custom, with an Evaluation tier for newer accounts. Each tier sets a monthly spend cap plus per-model request and token limits.

This article is about the ladder itself: how it is structured, what moves an organization up, what each rung changes, where to see your tier, and what to do when you outgrow it. The measurement side, including the three limit axes, response headers and retry logic, is covered in our guide to how Anthropic API rate limits are measured, so it is not repeated here.

The Anthropic usage tier ladder at a glance

Anthropic's rate limits page describes two kinds of limit, and both are defined by your tier:

• Spend limits set a maximum monthly cost your organization can incur on the API.
• Rate limits set the maximum number of requests and tokens your organization can use over a defined period.

Both are set at the organization level, not per API key. These are the tiers named on the official page at the time of writing:

If you searched for "tier 1" or "tier 4". Plenty of search queries and third-party guides refer to numbered tiers. The current official page does not use numbered names, so treat any numbered-tier figures you find elsewhere as possibly out of date, and check them against the live page before you plan capacity around them.

What moves an organization up a tier

The official page is brief on this point, and it is worth reading exactly rather than filling in gaps:

• Organizations are placed on a tier automatically, based on usage history and account standing.
• They can move to a higher tier over time as they use the API.
• Evaluation-tier limits increase automatically as the organization builds usage history. Anthropic describes these starting limits as part of how it prevents fraud and abuse.

At the time of writing, the rate limits page does not publish a deposit amount, a credit-purchase threshold or a waiting period that triggers a move. If you see thresholds like that quoted somewhere, they are not coming from the current official page, so verify them before relying on them. The documented routes are usage history and account standing, or an explicit request, which is covered further down.

In practice, a brand-new organization should not assume it will get the full Start-tier figures on day one. If you are planning a launch, run your load tests against the limits the Console actually shows for your organization, not against the table below.

One documented exception. Organizations using Claude Platform on AWS are placed on the Start tier and do not move between usage tiers automatically. Billing there runs through AWS Marketplace, and higher limits are requested through an Anthropic account representative or Anthropic support rather than the Console's increase flow.

What each Anthropic rate limit tier changes

The tables below reflect Anthropic's official rate limits page at the time of writing (September 2026). Limits change, so confirm the figures on the live page before sizing a production workload.

Monthly spend cap. Each of the Start, Build and Scale tiers carries a cap on what the organization can spend on the API in a calendar month:

Messages API limits per model. These are measured in requests per minute (RPM), input tokens per minute (ITPM) and output tokens per minute (OTPM), and they apply separately for each model. Several current models share identical figures, so they are grouped here:

Above Scale, the official page directs you to contact sales for a Custom arrangement. The footnotes on that page change how you should read the rows:

• Opus 4.x is one combined limit shared by Opus 4.8, 4.7, 4.6 and 4.5. Opus 5 has its own separate limit.
• Sonnet 4.x is one combined limit shared by Sonnet 4.6 and 4.5. Sonnet 5 has its own separate limit.
• Fable 5.x is one combined limit across Fable 5.1 and Fable 5. Mythos 5.1 and Mythos 5 share a separate combined limit on the same terms.
• Because limits are applied per model, you can run different models up to their respective limits at the same time.

Message Batches API limits. Batches have their own tier-based limits, shared across all models:

What sits outside the tier tables. The official page lists a few limits separately from the Start, Build and Scale tables: Claude Managed Agents endpoints have their own per-organization limits shown as a single set of values, the Files API has its own per-organization limit documented on the Files page, and fast mode has dedicated limits separate from the standard Opus limits. Do not assume those scale with your tier unless the relevant page says so.

How to check your current tier and limits in the Console

Everything tier-related lives in the Claude Console, spread across three screens:

1. Rate limits page (Settings > Limits). Shows your organization's tier and current limits. The Request rate limit increase option is here too.
2. Billing page (Settings > Billing). Shows the monthly spend cap for your tier and lets you set your own lower spend limit. Our walkthrough of the Anthropic API billing page explains the other figures on that screen.
3. Usage page. Besides token and request charts, it has two rate limit charts, one for input tokens and one for output tokens. They plot your hourly maximum per-minute usage against your current limit, and the input chart also shows your cache rate. This is the quickest way to see how much headroom your tier really leaves.

For automation, Anthropic also documents a Rate Limits API that reads configured organization and workspace limits programmatically. It is handy for a pre-launch check that compares configured limits with the load you plan to send.

When your tier is not enough

Start by confirming which ceiling you actually hit. The tier imposes two different ones, and they fail differently.

The tier's spend cap. Once an organization reaches its tier's monthly spend cap, usage pauses until 00:00 UTC on the first day of the next month, unless a higher limit is granted sooner. Requests return HTTP 429 with the error type rate_limit_error, but without a retry-after header. On the Messages API, the body carries a code that tells it apart from an ordinary rate limit. This is a trimmed version of the example on the official page:

{
  "type": "error",
  "error": {
    "type": "rate_limit_error",
    "details": { "error_code": "enforced_spend_limit_reached" }
  }
}

Retrying, including an SDK's automatic retries, keeps failing until access resumes. According to the official page, moving to a higher tier restores access.

A spend limit you set yourself. If you set your own limit below the tier cap, reaching it returns HTTP 400 with invalid_request_error instead. That is not a tier problem. Raise or remove your own limit.

A per-minute rate limit. A 429 that does carry retry-after is a rate limit, and the handling belongs to the measurement guide linked above. For tier purposes, the only question is whether your steady peak genuinely exceeds your tier's figures, which the Usage page charts will answer.

If the tier really is the constraint, work through these options, cheapest first:

1. Use more of what the tier already gives you. According to the official page, for most current models only uncached input tokens count toward ITPM, so cached context stretches the same limit further. Our notes on prompt caching with the Claude API cover how to set that up. Work that can wait can go through the Message Batches API, which has its own limits. And since limits are per model, a secondary job on a model outside your busiest bucket does not draw from the same limit.
2. Ramp gradually. The page notes that a sharp rise in usage can trigger acceleration limits, which also return 429s, and recommends ramping traffic gradually and keeping usage consistent. This is worth planning for whenever a new feature goes live.
3. Request an increase. Use Request rate limit increase on the Console's Rate limits page. It covers both higher rate limits and a higher monthly spend cap. For urgent needs, Anthropic support can also raise limits.
4. Move to Custom. If you need more than the Scale tier, the official page directs you to contact sales through the same Rate limits page.

For the request itself, the official guidance for Claude Platform on AWS asks for the models you need raised, your peak input and output tokens per minute for each, and roughly what share of your input is cached. That makes a sensible checklist for any increase request, and the Usage page gives you the numbers.

How tiers relate to workspace limits and rate limiting

The tier sets the organization-wide ceiling. Everything else sits underneath it:

• Workspace limits can be set per workspace, for spend and for each limiter type, but only lower than the organization's. Unset workspace limits match the organization's limits, the default workspace cannot have its own limits, and organization-wide limits always apply even if workspace limits add up to more.
• Enforcement uses a token bucket, so capacity replenishes continuously rather than resetting on the minute, and a per-minute figure can be enforced over shorter intervals. A higher tier gives you higher limits; the enforcement behaviour described on the page is the same.
• Limits are ceilings, not guarantees. The official page states that all limits represent maximum allowed usage, not guaranteed minimums.

If you reach Claude through ROIBest AI rather than directly, keep in mind that a gateway applies its own quota model, so compare its limits against the tier needs you worked out above.

FAQ

Where are Tier 1 to Tier 4 on Anthropic's rate limits page?

The official page at the time of writing does not use numbered tiers. It lists Evaluation, Start, Build, Scale and Custom, so check any numbered-tier figures from other sources against the live page.

How do I get to a higher Anthropic usage tier?

Anthropic places organizations on tiers automatically based on usage history and account standing. To move sooner, use Request rate limit increase on the Console's Rate limits page, or contact sales for limits above Scale.

Does setting my own spend limit change my tier?

No. Your own spend limit cannot exceed your current tier's cap, and reaching it returns an HTTP 400 that you clear by raising or removing that limit.

Are Anthropic tier limits applied per API key?

No. Limits are set at the organization level. You can set lower limits for individual workspaces, but organization-wide limits always apply.