> ## Documentation Index
> Fetch the complete documentation index at: https://docs.arouter.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Service Tiers

> Control the tradeoff between cost and latency using the service_tier parameter. ARouter passes service tier preferences to providers that support it.

The `service_tier` parameter lets you express a preference for how ARouter and the upstream provider should balance cost and latency for a request.

## Usage

```json theme={null}
{
  "model": "openai/gpt-5.4",
  "messages": [{"role": "user", "content": "Hello"}],
  "service_tier": "auto"
}
```

| Value              | Description                                                   |
| ------------------ | ------------------------------------------------------------- |
| `"auto"` (default) | Provider picks the appropriate tier based on availability     |
| `"default"`        | Standard tier — best cost-performance balance                 |
| `"flex"`           | Reduced cost, best-effort latency — ideal for batch workloads |

## Provider Support

Service tier is currently supported by:

| Provider | Supported values                | Notes                                                    |
| -------- | ------------------------------- | -------------------------------------------------------- |
| OpenAI   | `"auto"`, `"default"`, `"flex"` | `"flex"` offers reduced pricing for batch-like workloads |
| Others   | Ignored                         | Passed through but has no effect                         |

## Response

The `service_tier` used is echoed back in the response:

```json theme={null}
{
  "id": "gen-...",
  "model": "openai/gpt-5.4",
  "service_tier": "default",
  "choices": [...],
  "usage": {...}
}
```

## Use Cases

**Batch processing (cost-optimized):**

```python theme={null}
# Process large volumes at reduced cost
for document in documents:
    response = client.chat.completions.create(
        model="openai/gpt-5.4",
        messages=[{"role": "user", "content": f"Summarize: {document}"}],
        extra_body={"service_tier": "flex"},
    )
```

**Interactive applications (latency-optimized):**

```python theme={null}
# Real-time user-facing responses
response = client.chat.completions.create(
    model="openai/gpt-5.4",
    messages=[{"role": "user", "content": user_message}],
    extra_body={"service_tier": "default"},
)
```

## Related

* [Provider Routing](/en/provider-routing) — Fine-grained provider selection and throughput preferences
* [Latency and Performance](/en/guides/best-practices/latency-and-performance) — Best practices for low-latency applications