← Back to daily report

api/service-tiers.md

Changed on 2026-03-13 13:30:29 EST

+5 lines added
-5 lines removed
Visual Diff
# Service tiers¶

Different tiers of service allow you to balance availability, performance, and predictable costs based on your application's needs.¶

---¶

Anthropic offers three service tiers:¶
- **Priority Tier:** Best for workflows deployed in production where time, availability, and predictable pricing are important¶
- **Standard:** Default tier for both piloting and scaling everyday use cases¶
- **Batch:** Best for asynchronous workflows which can wait or benefit from being outside your normal capacity¶

## Standard Tier¶

The standard tier is the default service tier for all API requests. The API prioritizes these requests alongside all other requests with best-effort availability.¶

## Priority Tier¶

The API prioritizes requests in this tier over all other requests. This prioritization helps minimize ["server overloaded" errors](/docs/en/api/errors#http-errors), even during peak times.¶

For more information, see [Get started with Priority Tier](#get-started-with-priority-tier)¶

## How requests get assigned tiers¶

When handling a request, Anthropic decides to assign a request to Priority Tier in the following scenarios:¶
- Your organization has sufficient priority tier capacity **input** tokens per minute¶
- Your organization has sufficient priority tier capacity **output** tokens per minute¶

Anthropic counts usage against Priority Tier capacity as follows:¶

**Input Tokens**¶
- Cache reads as 0.1 tokens per token read from the cache¶
- Cache writes as 1.25 tokens per token written to the cache with a 5 minute TTL¶
- Cache writes as 2.00 tokens per token written to the cache with a 1 hour TTL¶
- For [long-context](/docs/en/build-with-claude/context-windows) (>200k input tokens) requests on Claude Sonnet 4.5 and Sonnet 4, input tokens are 2 tokens per token¶
- For [US-only inference](/docs/en/build-with-claude/data-residency) (`inference_geo: "us"`) requests on Claude Opus 4.6 and newer models, input tokens are 1.1 tokens per token¶
- All other input tokens are 1 token per token¶

**Output Tokens**¶
- For [long-context](/docs/en/build-with-claude/context-windows) (>200k input tokens) requests on Claude Sonnet 4.5 and Sonnet 4, output tokens are 1.5 tokens per token¶
- For [US-only inference](/docs/en/build-with-claude/data-residency) (`inference_geo: "us"`) requests on Claude Opus 4.6 and newer models, output tokens are 1.1 tokens per token¶
- All other output tokens are 1 token per token¶

Otherwise, requests proceed at standard tier.¶

<Note>¶
These burndown rates reflect the relative pricing of each token type. For example, US-only inference is priced at 1.1x on Opus 4.6 and newer models, so each token consumed with `inference_geo: "us"` draws down 1.1 tokens from your Priority Tier capacity. Multipliers stack: a long-context request with US-only inferenceFor Claude Sonnet 4.5 and Sonnet 4, long-context requests (>200k input tokens) draws down input tokens at 2.2 tokens per token (2 × 1.1)and output tokens at 1.5 tokens per token.¶
</Note>¶

<Note>¶
Requests assigned Priority Tier pull from both the Priority Tier capacity and the regular rate limits.¶
If servicing the request would exceed the rate limits, the request is declined.¶
</Note>¶

## Using service tiers¶

You can control which service tiers can be used for a request by setting the `service_tier` parameter:¶

```python¶
message = client.messages.create(¶
model="claude-opus-4-6",¶
max_tokens=1024,¶
messages=[{"role": "user", "content": "Hello, Claude!"}],¶
service_tier="auto", # Automatically use Priority Tier when available, fallback to standard¶
)¶
```¶

The `service_tier` parameter accepts the following values:¶

- `"auto"` (default) - Uses the Priority Tier capacity if available, falling back to your other capacity if not¶
- `"standard_only"` - Only use standard tier capacity, useful if you don't want to use your Priority Tier capacity¶

The response `usage` object also includes the service tier assigned to the request:¶

```json¶
{¶
"usage": {¶
"input_tokens": 410,¶
"cache_creation_input_tokens": 0,¶
"cache_read_input_tokens": 0,¶
"output_tokens": 585,¶
"service_tier": "priority"¶
}¶
}¶
```¶
This allows you to determine which service tier was assigned to the request.¶

When requesting `service_tier="auto"` with a model with a Priority Tier commitment, these response headers provide insights:¶
```text¶
anthropic-priority-input-tokens-limit: 10000¶
anthropic-priority-input-tokens-remaining: 9618¶
anthropic-priority-input-tokens-reset: 2025-01-12T23:11:59Z¶
anthropic-priority-output-tokens-limit: 10000¶
anthropic-priority-output-tokens-remaining: 6000¶
anthropic-priority-output-tokens-reset: 2025-01-12T23:12:21Z¶
```¶
You can use the presence of these headers to detect if your request was eligible for Priority Tier, even if it was over the limit.¶

## Get started with Priority Tier¶

You may want to commit to Priority Tier capacity if you are interested in:¶
- **Higher availability**: Target 99.5% uptime with prioritized computational resources¶
- **Cost Control**: Predictable spend and discounts for longer commitments¶
- **Flexible overflow**: Automatically falls back to standard tier when you exceed your committed capacity¶

Committing to Priority Tier will involve deciding:¶
- A number of input tokens per minute¶
- A number of output tokens per minute¶
- A commitment duration (1, 3, 6, or 12 months)¶
- A specific model version¶

<Note>¶
The ratio of input to output tokens you purchase matters. Sizing your Priority Tier capacity to align with your actual traffic patterns helps you maximize utilization of your purchased tokens.¶
</Note>¶

### Supported models¶

Priority Tier is supported by:¶

- Claude Opus 4.6¶
- Claude Opus 4.5¶
- Claude Opus 4.1¶
- Claude Opus 4¶
- Claude Sonnet 4.6¶
- Claude Sonnet 4.5¶
- Claude Sonnet 4¶
- Claude Sonnet 3.7 ([deprecated](/docs/en/about-claude/model-deprecations))¶
- Claude Haiku 4.5¶
- Claude Haiku 3.5 ([deprecated](/docs/en/about-claude/model-deprecations))¶

Check the [model overview page](/docs/en/about-claude/models/overview) for more details on available models.¶

### How to access Priority Tier¶

To begin using Priority Tier:¶

1. [Contact sales](https://claude.com/contact-sales/priority-tier) to complete provisioning¶
2. (Optional) Update your API requests to optionally set the `service_tier` parameter to `auto`¶
3. Monitor your usage through response headers and the Claude Console

Unified Diff

--- a/api/service-tiers.md
+++ b/api/service-tiers.md
@@ -31,19 +31,19 @@
 - Cache reads as 0.1 tokens per token read from the cache
 - Cache writes as 1.25 tokens per token written to the cache with a 5 minute TTL
 - Cache writes as 2.00 tokens per token written to the cache with a 1 hour TTL
-- For [long-context](/docs/en/build-with-claude/context-windows) (>200k input tokens) requests, input tokens are 2 tokens per token
-- For [US-only inference](/docs/en/build-with-claude/data-residency) (`inference_geo: "us"`) requests, input tokens are 1.1 tokens per token
+- For [long-context](/docs/en/build-with-claude/context-windows) (>200k input tokens) requests on Claude Sonnet 4.5 and Sonnet 4, input tokens are 2 tokens per token
+- For [US-only inference](/docs/en/build-with-claude/data-residency) (`inference_geo: "us"`) requests on Claude Opus 4.6 and newer models, input tokens are 1.1 tokens per token
 - All other input tokens are 1 token per token
 
 **Output Tokens**
-- For [long-context](/docs/en/build-with-claude/context-windows) (>200k input tokens) requests, output tokens are 1.5 tokens per token
-- For [US-only inference](/docs/en/build-with-claude/data-residency) (`inference_geo: "us"`) requests, output tokens are 1.1 tokens per token
+- For [long-context](/docs/en/build-with-claude/context-windows) (>200k input tokens) requests on Claude Sonnet 4.5 and Sonnet 4, output tokens are 1.5 tokens per token
+- For [US-only inference](/docs/en/build-with-claude/data-residency) (`inference_geo: "us"`) requests on Claude Opus 4.6 and newer models, output tokens are 1.1 tokens per token
 - All other output tokens are 1 token per token
 
 Otherwise, requests proceed at standard tier.
 
 <Note>
-These burndown rates reflect the relative pricing of each token type. For example, US-only inference is priced at 1.1x, so each token consumed with `inference_geo: "us"` draws down 1.1 tokens from your Priority Tier capacity. Multipliers stack: a long-context request with US-only inference draws down input tokens at 2.2 tokens per token (2 × 1.1).
+These burndown rates reflect the relative pricing of each token type. For example, US-only inference is priced at 1.1x on Opus 4.6 and newer models, so each token consumed with `inference_geo: "us"` draws down 1.1 tokens from your Priority Tier capacity. For Claude Sonnet 4.5 and Sonnet 4, long-context requests (>200k input tokens) draw down input tokens at 2 tokens per token and output tokens at 1.5 tokens per token.
 </Note>
 
 <Note>