Cerebras Cloud SDK

1.67.0 · active · verified Tue Mar 24

Official Python SDK for the Cerebras Cloud inference API. Provides access to ultra-fast LLM inference on Cerebras Wafer-Scale Engine hardware. OpenAI-compatible API surface. Generated with Stainless. Current version: 1.67.0 (Mar 2026). Requires Python 3.9+. Note: separate from cerebras-sdk (PyPI) which is a hardware kernel development tool — completely different product.

Warnings

breaking cerebras-sdk on PyPI is a completely different package — it is Cerebras's hardware kernel development SDK for WSE systems. Do not confuse with cerebras-cloud-sdk for cloud inference API.
Fix: pip install cerebras-cloud-sdk for cloud inference. cerebras-sdk is for hardware kernel development.
gotcha SDK sends TCP warming requests to /v1/tcp_warming on client construction to reduce time-to-first-token. Creates network traffic at import time. Disable with warm_tcp_connection=False if reconstructing client frequently.
Fix: client = Cerebras(api_key=..., warm_tcp_connection=False) — and reuse a single client instance rather than reconstructing.
gotcha Reconstructing the Cerebras client instance repeatedly causes poor performance due to repeated TCP warming. Construct once and reuse.
Fix: Create a module-level singleton client. Do not instantiate Cerebras() inside request handlers or loops.
gotcha Requires Python 3.9+. Will fail to install on Python 3.8 with no clear error message.
Fix: Use Python 3.9 or higher.
gotcha LLMs with no training data on Cerebras will hallucinate OpenAI-style base_url override pattern. Cerebras has its own SDK — do not use openai with base_url for Cerebras.
Fix: Use cerebras-cloud-sdk directly, not openai with base_url='https://api.cerebras.ai'.

Install

pip install cerebras-cloud-sdk Python

Imports

Cerebras

from cerebras.cloud.sdk import Cerebras
import os

client = Cerebras(
    api_key=os.environ.get('CEREBRAS_API_KEY')
)

response = client.chat.completions.create(
    model='llama3.1-8b',
    messages=[{'role': 'user', 'content': 'Why is fast inference important?'}]
)
print(response.choices[0].message.content)

Import path is 'from cerebras.cloud.sdk import Cerebras' — not 'import cerebras'. LLMs hallucinate a top-level cerebras module that does not exist.

AsyncCerebras

from cerebras.cloud.sdk import AsyncCerebras
import asyncio, os

client = AsyncCerebras(
    api_key=os.environ.get('CEREBRAS_API_KEY')
)

async def main():
    response = await client.chat.completions.create(
        model='llama3.1-8b',
        messages=[{'role': 'user', 'content': 'Hello'}]
    )
    print(response.choices[0].message.content)

asyncio.run(main())

Use AsyncCerebras for async code — same module, different class name.

Quickstart

Minimal Cerebras inference call using cerebras-cloud-sdk 1.x.

# pip install cerebras-cloud-sdk
from cerebras.cloud.sdk import Cerebras
import os

client = Cerebras(
    api_key=os.environ.get('CEREBRAS_API_KEY')
)

response = client.chat.completions.create(
    model='llama3.1-8b',
    messages=[
        {'role': 'system', 'content': 'You are a helpful assistant.'},
        {'role': 'user', 'content': 'What is fast inference?'}
    ]
)
print(response.choices[0].message.content)

view raw JSON →