is now an ernd partner: 10% back on inference 
In this guide
What are the tiers?
Cerebras has three: Free Trial, Developer (pay as you go) and Enterprise. For gpt-oss-120b, its docs list:
| Tier | Requests a minute | Uncached tokens a minute | Total tokens a minute |
|---|---|---|---|
| Free Trial | 5 | 30K | 90K |
| Developer | 1K | 1M | 3M |
| Enterprise | Set per organization | Set per organization | Set per organization |
The Free Trial also caps tokens per hour and per day, at 1M each for that model. Developer has no hourly or daily caps. Other models have their own numbers, so look yours up.
Cerebras calls this an overview where specific cases may vary. Your real limits are in the Cloud Console under Limits.
Your AI bill, with ernd
What could your AI bill earn?
Enter an amount from $0 to $100,000,000.
back a year at 4%
For US companies. No annual fee. Illustration based on AI vendor spend only. Rewards are points redeemable as cash back.
How do you move up?
- Buy credits (Pay as You Go) in the Cloud Console. Your first purchase moves the organization to Developer.
- Access comes back right away if your trial had stopped. Trial credits run out or expire after 30 days, and API access stops until you buy.
- For more than Developer, contact Cerebras or your account representative. Enterprise limits are provisioned per organization based on your usage profile.
How do you get more out of the limits you have?
Cerebras uses two token buckets. The uncached one counts cache misses, and Cerebras calls it the main constraint on throughput. The total one counts cached and uncached tokens together and defaults to three times the uncached limit. Either one can stop a request. A 429 tells you which bucket you went over.
So caching helps a lot. Cached tokens don’t touch the uncached limit, and the same limit then carries far more total traffic.
Set max_completion_tokens too. Cerebras estimates each request’s cost up front from your prompt plus that number, or its own guess at the output. Leave it high and requests get rate limited for tokens they’d never use.
The bucket refills continuously, so there’s no reset moment to wait for. Short backoffs usually beat waiting out a full minute.
Credits go on a card. ernd is a credit card for US companies with 4% back on AI vendor spend and 1.5% on everything else. Request the card at app.ernd.ai.
See how Cerebras compares with other vendors on our AI API rate limits page.
The ernd card
Keep capacity and card rewards separate.
The card earns rewards on AI spend. It does not bypass provider quotas, fix account restrictions or change provider prices.
Request your cardThe ernd card is a credit card for US companies, with its own partners in the AI stack. ernd isn't affiliated with any of the vendors on this page.
Common questions.
How much is the Cerebras Free Trial?
New accounts get $5 in credits after adding a verified payment method. The credits expire 30 days after they're granted.
Why do I still get 429s after buying Cerebras credits?
Buying credits moves you to the Developer tier, but Developer limits still apply per model. Check the Limits page in the Cloud Console for your numbers.
Do cached tokens count toward Cerebras limits?
Not toward the uncached limit. They do count toward the total limit, which defaults to three times your uncached limit.
Sources & methodology
Checked . Providers change prices and terms; the linked official page is the final word. This guide is published by ernd.
ernd is run by Provenance Research Inc.