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

# Inference service tiers

> Choose Flex, Default, or Priority for Telnyx Inference. Check model support, set service_tier, and handle unavailable capacity.

Service tiers select the serving capacity and pricing for a Telnyx-hosted model.
Set `service_tier` on each request and keep the same public `model` ID.

## Choose a tier

| Tier     | Request value                                          | When to use it                                                                                                                                                |
| -------- | ------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Flex     | <code style={{ whiteSpace: "nowrap" }}>flex</code>     | Cost-sensitive work that can tolerate higher latency and variable availability, such as offline evaluation, document processing, and background summarization |
| Default  | <code style={{ whiteSpace: "nowrap" }}>default</code>  | General-purpose inference with standard pricing                                                                                                               |
| Priority | <code style={{ whiteSpace: "nowrap" }}>priority</code> | Interactive applications where lower latency is more important than the lowest token price                                                                    |

For standard Inference API requests, omitting `service_tier` uses `default`.
Flex and Priority are available only on select models. A tier does not change the
model's capabilities or context window, and selecting Priority does not guarantee
a fixed response time.

Flex uses the same request APIs as the other tiers. A background workload sends
individual requests; selecting Flex does not create an asynchronous batch job.

## Check model support

Call [Get available models](/api-reference/openai-chat/get-available-models-openai-compatible)
(`GET /v2/ai/openai/models`) and inspect `data[].service_tiers` for the desired
model. Send one of those values as `service_tier`.

```bash theme={null}
curl https://api.telnyx.com/v2/ai/openai/models \
  -H "Authorization: Bearer $TELNYX_API_KEY"
```

Use the returned public `id` unchanged. Do not append a tier suffix to the model
name. Model support can change, so check the catalog before choosing Flex or
Priority. The `service_tiers` field is omitted for externally hosted models;
their provider determines supported tiers and behavior.

## Set the tier on a request

### Chat Completions

After confirming that `deepseek-ai/DeepSeek-V4.1-Flash` lists `flex`, send this
body to [Chat Completions](/api-reference/openai-chat/create-a-chat-completion-openai-compatible)
(`POST /v2/ai/openai/chat/completions`):

```json theme={null}
{
  "model": "deepseek-ai/DeepSeek-V4.1-Flash",
  "service_tier": "flex",
  "messages": [
    {
      "role": "user",
      "content": "Summarize this support note in one sentence: The customer could not sign in. Resetting the password restored access."
    }
  ]
}
```

To use Default or Priority, set `service_tier` to `default` or `priority` and
choose a model that advertises that tier. Streaming requests select the tier
with the same field.

### Responses

The [Responses API](/api-reference/openai-chat/create-an-openai-compatible-response)
(`POST /v2/ai/openai/responses`) accepts the same top-level field:

```json theme={null}
{
  "model": "deepseek-ai/DeepSeek-V4.1-Flash",
  "service_tier": "flex",
  "input": "Summarize this support note in one sentence: The customer could not sign in. Resetting the password restored access."
}
```

## Handle unavailable capacity

A supported tier can still be temporarily unavailable. For Flex workloads,
allow a longer request timeout and use bounded retries with exponential backoff
and jitter for transient capacity errors. Honor `Retry-After` when present.

If the model does not support the requested tier, select a value advertised by
the models endpoint instead of retrying the unchanged request. Do not rely on
automatic switching to another tier. An application can explicitly retry on
another supported tier if its latency requirements and budget permit; that new
request uses the newly selected tier's rates.

## Pricing

Input, cached input, and output token rates depend on both the model and service
tier. Compare the applicable rates on the
[Inference pricing page](https://telnyx.com/pricing/inference-api). See
[Pricing and billing units](/docs/inference/models/pricing) for the billing basis
of other Inference services.