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

# OpenAI compatibility

SambaNova inference APIs are designed to be compliant with OpenAI client libraries to simplify the adoption of SambaNova inference technologies to enhance your AI applications.

<Note>
  SambaNova also supports the Anthropic Messages API. See [Anthropic compatibility](/en/features/anthropic-compatibility) and the [Messages API](/en/features/messages) page.
</Note>

## Download the library

Run the command below to download the library.

```python theme={null}
pip install openai
```

## Use SambaNova APIs with OpenAI client libraries

Configure your OpenAI client libraries to use SambaNova inference APIs by setting two values: the `base_url` and your `api_key`.

<Note>
  Don’t have a SambaNova API key? Get yours from the [API keys and URLs](/en/get-started/api-keys-urls) page.
</Note>

```python theme={null}
from openai import OpenAI

client = OpenAI(
    base_url="your-sambanova-base-url",
    api_key="your-sambanova-api-key"
)
```

Now you can make an API request to a model and choose how to receive your output.

### Non-streaming example

The following code demonstrates using the OpenAI Python client for non-streaming completions.

```python theme={null}
completion = client.chat.completions.create(
  model="Meta-Llama-3.3-70B-Instruct",
  messages = [
      {"role": "system", "content": "Answer the question in a couple sentences."},
      {"role": "user", "content": "Share a happy story with me"}
    ]
)

print(completion.choices[0].message)
```

### Streaming example

The following code demonstrates using the OpenAI Python client for streaming completions.

```python theme={null}
completion = client.chat.completions.create(
  model="Meta-Llama-3.3-70B-Instruct",
  messages = [
      {"role": "system", "content": "Answer the question in a couple sentences."},
      {"role": "user", "content": "Share a happy story with me"}
    ],
  stream= True
)

for chunk in completion:
  print(chunk.choices[0].delta.content)
```

<Note>
  In streaming mode, the API returns chunks that contain multiple tokens. When calculating metrics like tokens per second or time per output token, count all tokens in each chunk.
</Note>

## Responses API

In addition to the Chat Completions endpoint, SambaNova exposes a Responses API endpoint `POST /v1/responses` that is compatible with the OpenAI Responses API standard. If you already use the OpenAI SDK with the Responses API, you can point it to SambaNova with the same configuration.

```python theme={null}
response = client.responses.create(
    model="gpt-oss-120b",
    input="Explain the difference between supervised and unsupervised learning in two sentences."
)

print(response.output[0].content[0].text)
```

For full details on supported parameters, tool calling, streaming, and structured output with the Responses API, see [Responses API](/en/features/responses).

## Currently unsupported OpenAI features

The following features are not yet supported and will be ignored:

* `presence_penalty`
* `frequency_penalty`

## Feature differences

`n`: The SambaNova API supports values 1–8 (integer, default 1). OpenAI has no documented hard cap. Note that `n` greater than 1 is not supported when using function calling or tools — combining them returns a 400 error.

`seed`: Supported on both SambaStack and SambaCloud for text generation models. Passing the same `seed` with the same inputs produces deterministic outputs. Accepts any integer, including negative values.

<Note>
  Unlike OpenAI, the SambaNova API does not return a `system_fingerprint` field in the response. The `seed` parameter is not supported for multi-modality models or continuous batching (CB) models.
</Note>

### `logit_bias`

Supported on both SambaStack and SambaCloud for text generation models. Accepts a JSON object mapping token IDs to bias values between `-100` and `100`. Negative values discourage the corresponding tokens; positive values encourage them. Token IDs are model-specific — look them up using the target model's tokenizer.

<Note>
  The `logit_bias` parameter is not supported for multimodal models or high throughput models. Requests with `logit_bias` against unsupported models return an error rather than silently ignoring the parameter.
</Note>

<Note>
  `logit_bias` is available only on deployments that support logit masking. Supported SambaStack 1.1.1 shipped models include this capability by default.
</Note>

## SambaNova API features not supported by OpenAI clients

The SambaNova API supports the `top_k` parameter, which is not supported by the OpenAI client libraries.