What Is Cerebras Inference? Free API Key, Trial Credit, Limits, Pricing, Models and Data

You send requests to Cerebras Inference from the Playground in your browser or from code through the API; it runs open-weight AI models very fast on Cerebras's cloud and returns the response. There's no permanent free tier: when you add a verified payment method, you get a one-time $5 credit that expires after 30 days; after that it's pay-as-you-go (per token). There are currently two public models: gpt-oss-120b and qwen-3.8-27b. The pricing page says the Developer tier is for development and testing, and that production workloads require Enterprise. The official policy states that inputs and outputs aren't used for model training. There's no official source on Turkish.

What it does: API that runs open models very fast
Free: Not permanently; one-time $5 credit, 30 days
Production: The Developer tier isn't for production
Training: Inputs and outputs aren't used for training
Turkish: Not stated in the official documentation
Price: gpt-oss-120b: $0.35 input / $0.75 output (1M tokens)

What Is Cerebras Inference Used For? Can You Use It Without Coding?

Cerebras Inference is an API service that lets you run a ready-made AI model on Cerebras's cloud without setting up your own server. An API is how one program sends requests to another: you send some text and an API key (a secret password that identifies your account), and the model sends back a response. In the page's terms, “inference” means a trained model producing a response. Cerebras's own model pages state these speeds: about 3000 tokens per second (gpt-oss-120b) and about 1850 tokens per second (qwen-3.8-27b). A token is roughly a piece of a word. These speeds are the vendor's own claims.

Getting fast responses from a modelWhat you get: Text responseNeeds a separate setup No-code Playground trialWhat you get: Model response, token count and speedBuilt into the app Asking questions with image inputWhat you get: Text responseNeeds a separate setup

Can I use it without coding?

Partly. The Playground in Cloud Console lets you try models from the browser without writing code. But for the free trial credit you need to add a verified payment method; if you don't, the Playground and API stay inactive. Connecting it to your app requires code.

Who it's for: developers who want to run models fast; people building apps where latency matters, such as voice agents, coding assistants or long-document summarization.

How Do You Get a Cerebras API Key?

Go to cloud.cerebras.ai (Cloud Console); the official docs say “sign up or log in” and don't specify the sign-up method (email, Google, etc.).
Add a verified payment method for the free trial credit. If you don't, the Playground and API stay inactive; you aren't charged for the credit.
Go to the API Keys page from the left menu and create a key. Free trial accounts have no project selection; you create the key directly.
Put the key in an environment variable (tabs below); don't write it into your code or share it with anyone.

Set the key› export CEREBRAS_API_KEY="your-key-here"

Set the key› $env:CEREBRAS_API_KEY = "your-key-here"

To make it permanent, the official docs recommend the setx CEREBRAS_API_KEY "…" command and opening a new terminal.

Python library› pip install --upgrade cerebras_cloud_sdk

Keys don't expire; if you delete one, it becomes invalid immediately and permanently, and apps using it get 401 Unauthorized. Whether phone verification, an age or a country requirement applies is not stated in the official documentation; see what the sign-up screen asks for.

How Do You Send Your First Request? A Ready-Made Example

The official quickstart shows how to send a message to the chat.completions endpoint. The Python example below is a simplified version of the code on that page (model names change often; current names are in the “Which models are available?” section):

Python

› import os
from cerebras.cloud.sdk import Cerebras

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

chat_completion = client.chat.completions.create(
    messages=[{"role": "user", "content": "Why is fast inference important?"}],
    model="qwen-3.8-27b",
)
print(chat_completion.choices[0].message.content)

curl (terminal)

› curl https://api.cerebras.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${CEREBRAS_API_KEY}" \
  -d '{"model": "qwen-3.8-27b", "messages": [{"role": "user", "content": "Why is fast inference important?"}]}'

No code: try it in the Playground"Summarize the customer message below in three bullet points, then write a short, polite draft reply: “My order still hasn't shipped after 5 days, and I want to cancel it.”"On the Cloud Console › Playground page, choose a model, paste the message and click Run. The response speed and token count appear at the top right. The reply can be in Turkish; there's no official benchmark for its quality.

The official docs say it also works with the OpenAI client libraries (the “OpenAI Compatibility” page); if you want to point existing OpenAI code at Cerebras, follow the steps there.

Error	What to do, according to the official docs
401 Unauthorized	`CEREBRAS_API_KEY` isn't set in the terminal where you ran the command. Check its value in the same terminal; on Windows, open a new terminal after `setx`.
404 model not found	The model name is wrong, deprecated or not available on your account. See the current list in the “Which models are available?” section.
429 Too Many Requests	You've hit a rate limit; the error message tells you whether the uncached or the total token limit was exceeded. Keep `max_completion_tokens` only as high as you need.

Which Models Does Cerebras Have? Which One Should You Choose?

Model (API name)	Cerebras's description	Speed	Context (free / paid)	Price: input / output (1M tokens)
OpenAI GPT OSS (`gpt-oss-120b`)	Efficient reasoning in science, math and coding; real-time coding help, long-document Q&A and summaries, agentic research	~3000 tokens/s	65k / 131k	$0.35 / $0.75
Qwen 3.8 27B (`qwen-3.8-27b`)	Agentic coding, tool use, research; text and image input, adjustable reasoning	~1850 tokens/s	64k / 128k	$0.99 / $1.49

Which one to choose?

Text only; speed and low price matter → gpt-oss-120b (summaries, Q&A, coding help).
You need image input (photos, screenshots) → qwen-3.8-27b (accepts Base64 PNG or JPEG; 2 images per request on free, 10 on Developer).
Thinking setting: reasoning_effort on gpt-oss (default medium); on Qwen, reasoning is high by default and turns off with none.

Model names in older examples may not work: gemma-4-31b has not been on Shared Inference since September 3, 2026; llama3.1-8b and qwen-3-235b-a22b-instruct-2507 were deprecated on May 27, 2026, and zai-glm-4.7 on August 17, 2026. Cerebras recommends these two models as replacements for the removed ones (it doesn't give a recommendation for zai-glm-4.7). More models (such as Kimi, GLM, MiniMax, Gemma) are available only on Dedicated Inference, by agreement.

The model list, pricing and speeds change often; this section was checked on October 15, 2026. Speed and quality statements are the vendor's own claims; there's no official source on the quality of Turkish replies.

Is Cerebras Inference Free? Trial Credit, Limits and Pricing

Free Trial$0 $5 credit, 30 days

One-time $5 credit when you add a verified payment method
All public models, with trial limits
You aren't charged for the credit
No permanent free tier
The credit expires after 30 days or when used up; the API and Playground stop
Hourly and daily token caps apply (1M a day)

Developer (pay-as-you-go)Per token

You move to it with your first credit purchase
No hourly or daily token caps
1,000 requests per minute on gpt-oss-120b
Pricing page: for development, evaluation and testing; not designed for production

EnterpriseContact sales

Production workloads
Higher limits and priority
Custom model weights and fine-tuning (by agreement)

Rate limits (checked October 15, 2026)

Model	Trial: requests/min	Trial: tokens/day	Developer: requests/min	Developer: tokens/min (uncached / total)
gpt-oss-120b	5	1M	1,000	1M / 3M
qwen-3.8-27b	5	1M	300	150K / 750K

Limits apply at the organization level and vary by model. Cached tokens don't count toward the uncached limit. Your exact values are on the Limits page in the Cloud Console left menu. The text on the gpt-oss-120b model page says 30 requests per minute for the trial; the limits page says 5: the two pages conflict.

Model prices are in the table in the “Which models are available?” section. Prices are in dollars; lira pricing, tax treatment and payment terms from Turkey aren't in the official sources. Credit is bought from the Billing tab; auto-reload is off by default.

The pricing page's footnote: the Developer tier “is intended for development, evaluation and testing; it is not designed for production use.” If you're planning a production workload that will serve real users, talk to Cerebras (Enterprise).

Does Cerebras Use Your Data for Training? Deleting Your Account

Training

Cerebras's policy page states that inputs and outputs from its AI services aren't stored, shared or used for model training. The Playground docs also say Playground and API requests are never used to train models. The terms of use separately state that they grant no right to train or fine-tune on service content.

Retention

Data is deleted once it's no longer needed to provide the service; no number of days is given. Account data such as name, email and payment details is collected separately.

Region and KVKK

Data may be processed in the US or other countries; Cerebras says it uses standard contractual clauses for this. There's no Turkey-specific text on KVKK (Turkey's Personal Data Protection Law). This isn't legal advice; if you'll process personal data, read the terms yourself.

Deleting your account

There's no self-service account deletion in the console. To delete it, you need to write to support@cerebras.net. Deleting an API key or archiving a project doesn't delete the account.

Frequently asked questions

How do you get a Cerebras API key, and is it free?

If you go to cloud.cerebras.ai and add a verified payment method, you get a one-time $5 credit that's valid for 30 days. Then you create a key on the API Keys page. There's no permanent free tier; when the credit runs out or expires after 30 days, the API and Playground stop.

What is Cerebras Inference, and why is it fast?

An API service that runs open models on Cerebras's cloud. On its own model pages, Cerebras states about 3000 (gpt-oss-120b) and 1850 (qwen-3.8-27b) tokens per second; these are the vendor's own figures.

How much does Cerebras Inference cost?

Pay-as-you-go, per million tokens: for gpt-oss-120b, $0.35 input and $0.75 output; for qwen-3.8-27b, $0.99 input and $1.49 output. Prices are in dollars; there's no lira pricing.

What are the Cerebras API limits, and why do you get a 429 error?

On the free trial, the official limits page says 5 requests per minute and 1 million tokens a day for gpt-oss-120b and Qwen; on Developer, gpt-oss-120b gets 1,000 requests per minute and no daily cap. A 429 means one of these limits was exceeded. Your exact values are on the Limits page in the Cloud Console.

Which models does Cerebras have?

There are currently two public models on Shared Inference: gpt-oss-120b (text) and qwen-3.8-27b (text and image input). More are available only on Dedicated Inference, by agreement.

Does Cerebras use my inputs for model training?

No: the official policy page states that inputs and outputs aren't used for model training. There's no button in the console to delete your account; you need to write to support@cerebras.net.

Related guides

Kimi

Azure AI Speech

Cloudflare Workers AI

Cohere API

Deepgram EMA Lightning

Sources

Not clear in the official sources: Turkish replies, interface and voice · sign-up requirements (phone, age, country; sign-up method) · conflicting trial limits (limits page 5, model page 30 requests/min) · payment and access from Turkey · lira pricing and tax · data retention period · Enterprise pricing. Each one is covered in its section, along with what you can do.