Cerebras Inference

cerebras.ai · Compute, Inference · 3 records · JSON

Cerebras Inference ComputetrialcardcardDretention-clamped
Cerebras InferencetrialcardcardDretention-clamped
Cerebras Inference Research Computefails: identityinstitutionno cardDwrite-gated

Cerebras Inference Compute

trialcardcardDretention-clamped

inference-docs.cerebras.ai/support/rate-limits · inference API · compute/cerebras-inference · JSON · compare

Free tier	No recurring free allowance. The FAQ reads: 'Is there a permanently free tier? No. The Free Trial is time- and credit-bounded: $5 in credits that expire 30 days after they are granted.' The Free Trial row that carries the eye-catching numbers (gpt-oss-120b and qwen-3.8-27b at 5 RPM, 30K uncached TPM, 90K total TPM, 1M TPH, 1M TPD) sits inside that 30-day, card-gated window, and the same page states 'New accounts receive $5 in free credits after adding a verified payment method.' The trap is that the per-model pages have not caught up: the gpt-oss-120b model page still prints 'Free tier rate limits: 30 requests/min, 60k input tokens/min, 1M tokens/day' as though a free tier existed. unit requests; metered on wall-clock; trial-credit
Price	none stated
Per-object cap	not applicable
Retention	the 30-day credit expiry binds before any rate does: USD 5 at gpt-oss-120b paid output pricing of USD 0.75 per 1M tokens is about 6.7M output tokens in total, or about 3.8M at a 3:1 input-to-output mix, so the whole allowance is one medium batch and not a monthly budget.
Egress	not stated
Identity	card card: a payment method is mandatory before any access - 'If you skip adding a payment method at sign-up, Playground and API access remain inactive until you do' - and the USD 5 arrives only 'after adding a verified payment method'. The pricing footnote adds 'You will not be charged or automatically enrolled in paid usage', so no authorisation or hold amount is stated in the evidence. Whether debit or prepaid is accepted is not stated: unknown.
Limits	retention-clamped the 30-day credit expiry binds before any rate does: USD 5 at gpt-oss-120b paid output pricing of USD 0.75 per 1M tokens is about 6.7M output tokens in total, or about 3.8M at a 3:1 input-to-output mix, so the whole allowance is one medium batch and not a monthly budget. source the 30-day credit expiry binds before any rate does: USD 5 at gpt-oss-120b paid output pricing of USD 0.75 per 1M tokens is about 6.7M output tokens in total, or about 3.8M at a 3:1 input-to-output mix, so the whole allowance is one medium batch and not a monthly budget.
Binding	retention-clamped the 30-day credit expiry binds before any rate does: USD 5 at gpt-oss-120b paid output pricing of USD 0.75 per 1M tokens is about 6.7M output tokens in total, or about 3.8M at a 3:1 input-to-output mix, so the whole allowance is one medium batch and not a monthly budget.
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	nothing composes into a recurring job: the 1M-tokens-a-day ceiling is real for 30 days and then access pauses until credits are purchased.
Reliability	no reliability event recorded
Notes	Corrects the shape of a claim that circulates widely: '1M tokens a day free on Cerebras'. The number is a daily rate limit inside a card-gated 30-day trial, so the monthly figure is not 30M tokens. Arithmetic, labelled estimate: USD 5 / (USD 0.75 per 1M output) = 6.67M output tokens if the credit were spent on output alone; at a 3:1 input-to-output split, USD 5 buys about 3.8M tokens of mixed traffic. The contradiction between the FAQ and the model pages is the real finding for a future reader: a session that reads only the model page records a free tier that the FAQ denies in as many words. See also the SambaNova record in a10-repos, the same pattern of a stale sub-page describing a withdrawn free tier.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (5)

date	source	what it proved
2026-10-22	inference-docs.cerebras.ai/support/rate-limits	FAQ answer 'Is there a permanently free tier? No. The Free Trial is time- and credit-bounded: $5 in credits that expire 30 days after they are granted', quoted
2026-10-22	inference-docs.cerebras.ai/support/rate-limits	FAQ answer 'New accounts receive $5 in free credits after adding a verified payment method. These credits expire 30 days after they are granted', quoted; and 'If you skip adding a payment method at sign-up, Playground and API access remain inactive until you do'
2026-10-22	inference-docs.cerebras.ai/support/rate-limits	Free Trial table: gpt-oss-120b and qwen-3.8-27b at 5 RPM, 30K uncached TPM, 90K total TPM, 1M TPH, 1M TPD, quoted
2026-10-22	www.cerebras.ai/pricing	pricing page footnote '*** Add a valid payment method to receive a one-time $5 promotional credit. The credit expires 30 days after activation. You will not be charged or automatically enrolled in paid usage.', quoted
2026-10-22	inference-docs.cerebras.ai/models/openai-oss	model page still prints 'Free tier rate limits: 30 requests/min, 60k input tokens/min, 1M tokens/day' alongside paid pricing of USD 0.35 input and USD 0.75 output per 1M tokens, which contradicts the FAQ's 'there is no permanently free tier'

More fields (1)

{
 "wave2": {
  "class": "refutation",
  "free_tier_kind": "trial-credit",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": "wafer-scale WSE",
   "gpu_count": null
  }
 }
}

Unknown or withheld values (7)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.tooling not-stated

Cerebras Inference

trialcardcardDretention-clamped

inference-docs.cerebras.ai/support/rate-limits · inference endpoint, keyed free trial · inference/cerebras · JSON · compare

Free tier	a Free Trial, not a free tier: USD 5 of credits granted after a verified payment method is added, expiring 30 days after they are granted. The page answers 'Is there a permanently free tier?' with 'No.' unit none; metered on request-count; the trial's per-model limits are published: gpt-oss-120b at 5 RPM, 30K uncached TPM, 90K total TPM, 1M TPH and 1M TPD; qwen-3.8-27b at 1 RPM, 5 uncached TPM and the same hourly and daily caps
Price	none stated
Per-object cap	not applicable: an inference endpoint stores nothing for the caller
Retention	the trial credits expire 30 days after they are granted (30 days)
Egress	the response is the payload; no separate egress term is published
Identity	card a verified payment method is required before the USD 5 of trial credits is granted, which makes this a card gate rather than an email gate
Limits	request-rate Free Trial, gpt-oss-120b: 5 RPM, 30K uncached TPM, 90K total TPM, 1M TPH, 1M TPD. qwen-3.8-27b: 1 RPM, 5 uncached TPM, same hourly and daily caps source retention-clamped the USD 5 of trial credits expire 30 days after they are granted, and API access stops when they run out or expire source the Developer tier removes the hourly and daily token caps and raises the per-model rates; that tier is pay-as-you-go and not free
Binding	retention-clamped the 30-day credit expiry binds before any rate does: the whole allowance is USD 5 that disappears, so there is no recurring quota to design against
Interface	rest an OpenAI-compatible chat completions endpoint at https://api.cerebras.ai/v1 endpoint: https://api.cerebras.ai/v1/chat/completions
Tooling	any OpenAI-compatible client; curl against POST {base}/chat/completions
Composes with	pairs with an agent only inside a 30-day trial and only after a card; the scarce link is the verified payment method, which turns a free offer into a payment relationship
Reliability	an unauthenticated call returns 401 with 'Wrong API Key' and a 99-byte body, so an unauthenticated probe cannot distinguish a live endpoint from an expired one
Notes	The vendor's own page says there is no permanently free tier: the offer is USD 5 of credits that expire in 30 days and require a verified payment method. The identity cost is therefore card, not email, and the verdict is trial-only.
Evidence rung	documented the trial's terms, its expiry, its per-model rates and the vendor's explicit 'No' to a permanently free tier are all quoted from the rate-limits page read 2026-10-22 at 200 with 393345 bytes. The 401 gate was also re-probed at 2026-10-22T04:42:43Z. No account was created, so the trial is documented and not measured.
Checked	2026-10-23 (live); stale on 2026-11-22

Evidence (2)

date	source	what it proved
2026-10-22	inference-docs.cerebras.ai/support/rate-limits	'New accounts receive $5 in free credits after adding a verified payment method. These credits expire 30 days after they're granted'; 'Is there a permanently free tier? No. The Free Trial is time- and credit-bounded'; and the Free Trial table giving gpt-oss-120b 5 RPM, 30K uncached TPM, 90K total TPM, 1M TPH and 1M TPD
2026-10-22	probe:https://api.cerebras.ai/v1/chat/completions	an unauthenticated POST returned 401 with a 99-byte body, {"message":"Wrong API Key","type":"invalid_request_error","param":"api_key"}. The endpoint exists and gates on a key; no key existed here

Cerebras Inference Research Compute

fails: identityinstitutionno cardDwrite-gated

www.cerebras.ai/blog/grantfrp · inference API · compute/cerebras-inference-research · JSON · compare

Free tier	'Selected Principal Investigators (PIs) may receive the following: Cerebras Inference service credits, equivalent to no more than $50,000 USD of inference credits' plus 'tutorials and hands-on sessions with Cerebras engineers and researchers'. The call is addressed to 'university faculty and researchers'. The amount is capped by the vendor and 'The final grant amount and details will be determined by Cerebras.' unit none; metered on wall-clock; grant-credits
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	institution not published
Limits	write-gated a written proposal reviewed by Cerebras, with a principal investigator attached to a university, is the gate; the currency is inference tokens rather than machine time, so no node-hour figure exists to plan against. source a written proposal reviewed by Cerebras, with a principal investigator attached to a university, is the gate; the currency is inference tokens rather than machine time, so no node-hour figure exists to plan against.
Binding	write-gated a written proposal reviewed by Cerebras, with a principal investigator attached to a university, is the gate; the currency is inference tokens rather than machine time, so no node-hour figure exists to plan against.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	It would compose with the free Cerebras inference tier already recorded, because both spend one account. It does not compose for an individual: the PI requirement is the wall, and the USD 50,000 is the largest single number that a person without an affiliation cannot touch.
Reliability	no reliability event recorded
Notes	The rung is documented, and the number is deliberately recorded with its ceiling wording. What is unknown and left null: the application window, whether the RFP is still open, and the token-to-dollar rate used to denominate the award. A reader should treat this as a closed door with a big number behind it, not as capacity.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (3)

date	source	what it proved
2026-10-22	www.cerebras.ai/blog/grantfrp	'Cerebras Inference service credits, equivalent to no more than $50,000 USD of inference credits', quoted, in the Grant section
2026-10-22	www.cerebras.ai/blog/grantfrp	'We invite university faculty and researchers to respond to this Request for Proposals (RFP)', quoted
2026-10-22	www.cerebras.ai/blog/grantfrp	'The final grant amount and details will be determined by Cerebras', quoted, so the USD 50,000 is a ceiling and not an award

More fields (1)

{
 "wave2": {
  "class": "institution-grant",
  "free_tier_kind": "grant-credits",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": "Cerebras Wafer Scale Engine 3, inference only",
   "gpu_count": null
  }
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated