Cerebras Inference Free Trial: $5 Credit & Rate Limits (2026)

Short answer: Cerebras does not have a permanent free tier anymore. New accounts get a one-time $5 credit once they add a verified payment method. The credit lasts 30 days, works on all public models, and comes with tight rate limits (5 requests per minute, 1 million tokens per day per model). You are not charged unless you buy more credits. It is a good way to see how fast Cerebras is, but not a way to run anything for free long term.

Checked against Cerebras's official rate-limits docs, models overview and pricing page on 2026-10-17. We list this offer as Limited because it needs a card and ends after 30 days.

What you get with the $5 credit

Cerebras's rate limits page describes the trial in plain terms: new accounts receive $5 in free credits after adding a verified payment method; the credits expire 30 days after they are granted and can be used across all public models; and there is no charge until you choose to buy more credits.

• $5 of usage on Cerebras's shared (serverless) inference API, billed per token from the credit.
• Access to all public models. On 2026-10-17 that meant OpenAI's gpt-oss-120b and Qwen 3.8 27B (qwen-3.8-27b), which also accepts images.
• Playground and API. You can test in the browser Playground or call the API with a key.
• No automatic charge. Cerebras says it does not bill you until you choose to purchase credits.

What you do not get

• No renewing allowance. Cerebras states there is no no-cost tier that renews automatically and no per-model always-free allowance.
• No production use. The pricing page marks the tier as for development, evaluation and experimentation, not production.
• No full context window. The free tier gets about half the paid context (details below).
• No Dedicated Inference. Other model families run on Dedicated Inference, which is arranged by agreement with Cerebras, not through the trial.

Free trial rate limits

The trial has five separate limits per model. Whichever one you hit first stops you.

• Uncached vs total tokens: prompt tokens that hit Cerebras's prompt cache count toward the total limit but not the uncached one. Reusing the same long system prompt lets you send more per minute.
• Images: on the trial, Qwen 3.8 27B accepts up to 2 images per request and a 10 MiB total request payload.
• Hourly and daily caps are the same (1M). You can use the whole day's allowance in one hour, then you are done until the daily window resets.

Context window: free vs paid

Figures from Cerebras's models overview. The speed is what makes Cerebras interesting: when it launched gpt-oss-120b in 2025, Cerebras said it ran at about 3,000 tokens per second, many times faster than typical GPU providers.

Who can claim it

• New Cerebras Cloud accounts. The credit is described as one-time and for new accounts.
• A verified payment method is required. If you skip it at sign-up, Cerebras says Playground and API access stay inactive until you add one.
• No student, startup or company check. Cerebras does not publish any other eligibility rules or country restrictions for the trial beyond its general terms.
• One-time only. The credit is a single promotional grant; Cerebras does not describe any way to get it again on the same account.

How to claim the $5 credit, step by step

1. Go to cloud.cerebras.ai and create an account.
2. Add a valid payment method when asked. This activates the $5 promotional credit; you are not charged for adding it.
3. Open the Playground to try gpt-oss-120b or Qwen 3.8 27B in the browser.
4. Create an API key in the console and call the API. Cerebras's API follows the OpenAI chat completions format, so most OpenAI SDKs work with a changed base URL.
5. Keep an eye on the 30-day clock. Credits expire 30 days after they are granted, whether you used them or not.

curl https://api.cerebras.ai/v1/chat/completions \
  -H "Authorization: Bearer $CEREBRAS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-oss-120b", "messages": [{"role": "user", "content": "Explain wafer-scale chips in two sentences."}]}'

What happens after 30 days or $5

When the credit runs out or expires, Cerebras says API and Playground access stop on the Free Trial tier until you purchase credits. Nothing is charged automatically. If you do buy credits, a pay-as-you-go purchase reactivates access immediately and moves you to the Developer tier, which raises rate limits and removes the hourly and daily token caps.

On the Developer tier, Qwen 3.8 27B also accepts up to 10 images per request. Cerebras says you can use as many tokens as you need within your budget. Per-token prices are on the pricing page; we don't quote them here. Check the live page before estimating costs.

How to get the most from a $5, 30-day trial

• Plan a speed test, not a project. The trial's real value is measuring latency and throughput on your own prompts. Prepare a test set first, then run it in one session.
• Reuse your system prompt. Cached prompt tokens do not count toward the 30K uncached tokens-per-minute limit, so a fixed system prompt lets you push more through.
• Stay under 5 requests per minute. Add a delay of about 12 seconds between calls in scripts, or you will spend more time on rate-limit errors than on tests.
• Compare the same model elsewhere. gpt-oss-120b is also served by other providers, so you can run the same prompts on another host and compare speed and cost directly.
• Use the full 30 days only if you need it. The credit expires on a fixed date, so start testing soon after you add your card.

Is the trial worth adding a card for?

It depends on what you want to learn. Adding a card costs nothing by itself, and Cerebras says it will not charge you unless you buy credits. The trade-off is that your card is on file with another vendor, and the credit is small and short-lived.

• Worth it if you are choosing an inference provider and response speed matters to your product, for example voice agents, live coding assistants or chat interfaces where users wait for every token. Thirty days is enough to benchmark Cerebras against your current host.
• Worth it if you already use gpt-oss-120b or a Qwen model and want to see how much faster the same model runs on Cerebras hardware.
• Probably not worth it if you only want free tokens for a side project. Groq, the Gemini API and OpenRouter's free models give you an ongoing allowance with no card and no expiry date.
• Not worth it if you need a model Cerebras does not offer on its shared API. Only two public models were listed when we checked.

How Cerebras compares with other fast or free APIs

If you want fast open-model inference with no card, Groq is the closest alternative. If you are a startup that needs much larger credits, look at programs such as NVIDIA Inception or our guide on how to get free AI credits. For a full list of keys you can get without paying, see free AI API keys.

History of the offer

• 2024 to 2025: Cerebras launched its inference API with a free tier that did not need a card. Many third-party lists still describe that tier, often with a 1M tokens per day figure.
• 2025: Cerebras added OpenAI's gpt-oss-120b and promoted speeds of about 3,000 tokens per second.
• 2026: the standing free tier was replaced by the current one-time $5 credit with a required payment method and a 30-day expiry. Cerebras does not publish the exact date of the change.
• Now (October 2026): the docs state there is no permanently free tier, and the shared catalog lists gpt-oss-120b and Qwen 3.8 27B.

Common mistakes

• Signing up without a card and waiting for access. The credit and API access only activate once a payment method is added.
• Reading old guides. If a guide promises unlimited daily tokens with no card, it describes the old free tier.
• Assuming paid limits. The 65K context and 5 requests per minute catch people who test long documents or agent loops. Paid accounts get the full context and much higher limits.
• Letting the credit expire unused. The 30 days start when the credit is granted, not when you first call the API.