Chat Completions - Cerebras Inference

messages

object[]

required

A list of messages comprising the conversation so far.Note: System prompts must be passed to the messages parameter as a string. Support for other object types will be added in future releases.

model

string

required

Available options:

llama-4-scout-17b-16e-instruct
llama3.1-8b
llama-3.3-70b
llama-4-maverick-17b-128e-instruct (preview)
qwen-3-32b
qwen-3-235b-a22b-instruct-2507 (preview)
qwen-3-235b-a22b-thinking-2507 (preview)
qwen-3-coder-480b (preview)
deepseek-r1-distill-llama-70b (to be deprecated August 12, 2025)

deepseek-r1-distill-llama-70b are available in private preview. Please contact us to request access.

max_completion_tokens

integer | null

The maximum number of tokens that can be generated in the completion. The total length of input tokens and generated tokens is limited by the model’s context length.

response_format

object | null

Controls the format of the model response. The primary option is structured outputs with schema enforcement, which ensures the model returns valid JSON adhering to your defined schema structure.Setting to { "type": "json_schema", "json_schema": { "name": "schema_name", "strict": true, "schema": {...} } } enforces schema compliance. The schema must follow standard JSON Schema format with the following properties:

Show json_schema properties

response_format.json_schema.name

string

An optional name for your schema.

response_format.json_schema.strict

boolean

When set to true, enforces strict adherence to the schema. The model will only return fields defined in the schema and with the correct types. When false, behaves similar to JSON mode but uses the schema as a guide.

response_format.json_schema.schema

object

A valid JSON Schema object that defines the structure, types, and requirements for the response. Supports standard JSON Schema features including types (string, number, boolean, integer, object, array, enum, anyOf, null), nested structures (up to 5 layers), required fields, and additionalProperties (must be set to false).

Note: Structured outputs with JSON schema is currently in beta. Visit our page on Structured Outputs for more information.

Show JSON mode

Alternatively, setting to { "type": "json_object" } enables simple JSON mode, which ensures that the response is either a valid JSON object or an error response without enforcing a specific schema structure.Note that enabling JSON mode does not guarantee the model will successfully generate valid JSON. The model may fail to generate valid JSON due to various reasons such as incorrect formatting, missing or mismatched brackets, or exceeding the length limit.In cases where the model fails to generate valid JSON, the error response will be a valid JSON object with a key failed_generation containing the string representing the invalid JSON. This allows you to re-submit the request with additional prompting to correct the issue. The error response will have a 400 server error status code.Note that JSON mode is not compatible with streaming. "stream" must be set to false.Important: When using JSON mode, you need to explicitly instruct the model to generate JSON through a system or user message.

seed

integer | null

If specified, our system will make a best effort to sample deterministically, such that repeated requests with the same seed and parameters should return the same result. Determinism is not guaranteed.

stop

string | null

Up to 4 sequences where the API will stop generating further tokens. The returned text will not contain the stop sequence.

stream

boolean | null

If set, partial message deltas will be sent.

temperature

number | null

What sampling temperature to use, between 0 and 1.5. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. We generally recommend altering this or top_p but not both.

top_p

number | null

An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So, 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or temperature but not both.

tool_choice

string | object

Controls which (if any) tool is called by the model. none means the model will not call any tool and instead generates a message. auto means the model can pick between generating a message or calling one or more tools. required means the model must call one or more tools. Specifying a particular tool via {"type": "function", "function": {"name": "my_function"}} forces the model to call that tool.none is the default when no tools are present. auto is the default if tools are present.

tools

object | null

A list of tools the model may call. Currently, only functions are supported as a tool. Use this to provide a list of functions the model may generate JSON inputs for.Specifying tools consumes prompt tokens in the context. If too many are given, the model may perform poorly or you may hit context length limitations

Show properties

tools.function.description

string

A description of what the function does, used by the model to choose when and how to call the function.

tools.function.name

string

The name of the function to be called. Must be a-z, A-Z, 0-9, or contain underscores and dashes, with a maximum length of 64.

tools.function.parameters

object

The parameters the functions accepts, described as a JSON Schema object. Omitting parameters defines a function with an empty parameter list.

tools.type

string

The type of the tool. Currently, only function is supported.

user

string | null

A unique identifier representing your end-user, which can help to monitor and detect abuse.

logprobs

bool

Whether to return log probabilities of the output tokens or not.Default: False

top_logprobs

integer | null

An integer between 0 and 20 specifying the number of most likely tokens to return at each token position, each with an associated log probability. logprobs must be set to true if this parameter is used.

from cerebras.cloud.sdk import Cerebras
import os 

client = Cerebras(api_key=os.environ.get("CEREBRAS_API_KEY"),)

chat_completion = client.chat.completions.create(
    model="llama3.1-8b",
    messages=[
        {"role": "user", "content": "Hello!",}
    ],
)
print(chat_completion)

{
  "id": "chatcmpl-292e278f-514e-4186-9010-91ce6a14168b",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "Hello! How can I assist you today?",
        "role": "assistant"
      }
    }
  ],
  "created": 1723733419,
  "model": "llama3.1-8b",
  "system_fingerprint": "fp_70185065a4",
  "object": "chat.completion",
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 10,
    "total_tokens": 22
  },
  "time_info": {
    "queue_time": 0.000073161,
    "prompt_time": 0.0010744798888888889,
    "completion_time": 0.005658071111111111,
    "total_time": 0.022224903106689453,
    "created": 1723733419
  }
}

Endpoints