Groq

groq.com · Compute, Inference, Storage · 4 records · JSON

Groq Cloud Computecaveatemailno cardDrequest-rate
Groq Inferencecaveatemailno cardDrequest-rate
groq files api Storagefails: quotaemail?card?R
Hosted inference APIs (as a compute class) Computedeademailno cardRrequest-rate

Groq Cloud Compute

caveatemailno cardDrequest-rate

console.groq.com/docs/rate-limits · inference API · compute/groq-cloud · JSON · compare

Free tier	A free plan per model, printed as RPM / RPD / TPM / TPD: openai/gpt-oss-120b and openai/gpt-oss-20b at 30 requests a minute, 1,000 a day, 8,000 tokens a minute and 200,000 tokens a day, which works out to 6.0M tokens a month per model. Paid Developer rates for the same model are USD 0.15 per 1M input and USD 0.60 per 1M output at 1,000 RPM and 250K TPM. unit tokens; metered on wall-clock; recurring-daily-allowance
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	email not published
Limits	request-rate the 8,000 tokens a minute binds before the 200,000 a day: at 8K TPM the day's whole allowance is reachable in 25 minutes of continuous traffic, so the pool is spent by mid-morning and the rest of the day is idle unless the job is paced. source the 8,000 tokens a minute binds before the 200,000 a day: at 8K TPM the day's whole allowance is reachable in 25 minutes of continuous traffic, so the pool is spent by mid-morning and the rest of the day is idle unless the job is paced.
Binding	request-rate the 8,000 tokens a minute binds before the 200,000 a day: at 8K TPM the day's whole allowance is reachable in 25 minutes of continuous traffic, so the pool is spent by mid-morning and the rest of the day is idle unless the job is paced.
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	the daily pool is small but the latency is the point: 500 tokens a second on this model spends the 200K-token day in about 7 minutes of generation, so Groq is the cheap first pass and Cloudflare or DeepSeek takes the volume.
Reliability	no reliability event recorded
Notes	Arithmetic, labelled estimate: 200,000 tokens a day is 6.0M a month per model, and about 24M a month across the four free text models if every model is used. At the Developer rate that 6.0M is worth roughly USD 0.90 to USD 3.60 a month depending on the input/output split. GPU-equivalent, labelled estimate: 6.0M output tokens a month at 500 tokens a second is about 3.3 hours of continuous single-stream generation a month. Confirms the corpus record (dossier/inference/groq.json): the numbers are unchanged on 2026-10-22. Adds the capacity shape, which that record does not carry.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21
Parsed numbers	`$.limits[0].value` limit:request-rate 8000.0 tokens / minute

Evidence (3)

date	source	what it proved
2026-10-22	console.groq.com/docs/rate-limits	'openai/gpt-oss-120b \| 30 \| 1K \| 8K \| 200K' in the free plan rate limit table, quoted
2026-10-22	console.groq.com/docs/models	Developer plan pricing for gpt-oss-120b at $0.15 input and $0.60 output per 1M tokens with 250K TPM and 1K RPM, quoted
2026-10-22	console.groq.com/docs/rate-limits	'Rate limits apply at the organization level, not individual users. You can hit any limit type depending on which threshold you reach first', quoted

More fields (1)

{
 "wave2": {
  "class": "capacity-shape",
  "free_tier_kind": "recurring-daily-allowance",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": "LPU",
   "gpu_count": null
  }
 }
}

Unknown or withheld values (9)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.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

Groq Inference

caveatemailno cardDrequest-rate

console.groq.com/docs/rate-limits · inference endpoint, keyed free tier · inference/groq · JSON · compare

Free tier	a Free Plan per model, printed as RPM / RPD / TPM / TPD in the page's own units: openai/gpt-oss-120b and openai/gpt-oss-20b at 30 / 1K / 8K / 200K, qwen/qwen3.8-27b the same, and openai/gpt-oss-safeguard-20b at 3 / 1K / 2K / 200K. The page writes requests-per-day and tokens-per-day in thousands, so 1K is 1,000 and 200K is 200,000 unit tokens; metered on request-count; the served page carries two tabs, Free Plan Limits and Developer Plan Limits, and seven model rows in the Free tab. The numbers are in the HTML and were parsed from it, not read from a live call The page prints the cells as 1K, 8K and 200K; the expansions here are the page's own units written out, not a conversion The page prints 1K, 8K and 200K verbatim; expanding K to a thousand is a unit reading of the page's own suffix and not a conversion between units
Price	none stated
Per-object cap	not applicable: an inference endpoint stores nothing for the caller
Retention	not applicable
Egress	the response is the payload; no separate egress term is published
Identity	email an account and an API key
Limits	request-rate 30 / 1K / 8K / 200K on the Free plan for openai/gpt-oss-120b and openai/gpt-oss-20b, in the page's own units: 30 requests a minute, 1K requests a day, 8K tokens a minute and 200K tokens a day source request-rate RPM, RPD, TPM, TPD, ITPM and OTPM at organisation level; every dimension is evaluated and the first threshold reached binds. Cached tokens count toward none of them source
Binding	request-rate TPD binds on a user's workload: 200,000 tokens a day is 6.0M a month, which is 11.4 per cent of a 52.5M-token monthly target. The 30 rpm and 8,000 tpm figures are both workable for an interactive session
Interface	rest endpoint: https://api.groq.com/openai/v1/chat/completions
Tooling	any OpenAI-compatible client; curl against POST {base}/chat/completions
Composes with	pairs with a free-standing agent as the email-gated fallback when a keyless endpoint's ceiling binds; the 200,000-token daily cap is the scarce link, not the request rate
Reliability	no reliability event recorded; no key was available
Notes	Groq names six limit dimensions where most vendors name one, and it prints the free values where this topic's other keyed tiers do not. A pipeline sized on RPM alone meets TPD first: 200,000 tokens a day is 6.0M a month against a 52.5M-token target.
Evidence rung	documented the dimensions, the organisation scope and the seven-row free-plan table are all quoted from the served page read today. No key was available, so the enforcement of any of the numbers is not measured verified_at is the instant its evidence URLs were fetched on 2026-10-22, read from the fetch log, not typed.
Checked	2026-10-23 (live); stale on 2026-11-22
Parsed numbers	`$.limits[0].value` limit:request-rate 30.0 requests / minute

Evidence (1)

date	source	what it proved
2026-10-22	console.groq.com/docs/rate-limits	limits are RPM, RPD, TPM, TPD, ITPM and OTPM, applied at organisation level, evaluated against every dimension with the first reached binding; cached tokens do not count; the Free Plan Limits tab carries seven model rows, of which openai/gpt-oss-120b and openai/gpt-oss-20b are 30 rpm, 1K rpd, 8K tpm and 200K tpd The Free Plan Limits tab prints the cells as 30 / 1K / 8K / 200K, and the page's K suffix means thousands

Unknown or withheld values (1)

$.retention.stated not-applicable

groq files api Storage

fails: quotaemail?card?R

console.groq.com/docs/api-reference · rejected-provider · storage/groq-files-api · JSON · compare

Free tier	Batch-only purpose, .jsonl only, 100 MB max file. No general-purpose storage class. unit none; metered on none
Price	none stated
Per-object cap	not recorded
Retention	not recorded
Egress	not recorded
Identity	email (placeholder) not published
Binding	none no constraint of a named kind was recorded; the verdict rests on the quota sentence rather than on a named limit
Interface	unknown
Tooling	not recorded
Composes with	not recorded
Reliability	Batch-only purpose, .jsonl only, 100 MB max file. No general-purpose storage class.
Notes	Batch-only purpose, .jsonl only, 100 MB max file. No general-purpose storage class.
Evidence rung	reported
Checked	2026-10-21 (inherited); stale on 2026-11-04

Evidence (2)

date	source	what it proved
2026-10-21	console.groq.com/docs/api-reference	Batch-only purpose, .jsonl only, 100 MB max file. No general-purpose storage class.
2026-10-21	raw.githubusercontent.com/api-evangelist/groq/refs/heads/mai	cited as the basis of the rejection: Batch-only purpose, .jsonl only, 100 MB max file. No general-purpose storage class.

More fields (1)

{
 "dead_end": {
  "failure_class": "quota",
  "why": "Batch-only purpose, .jsonl only, 100 MB max file. No general-purpose storage class.",
  "retry_when": null,
  "first_recorded": "2026-10-21"
 }
}

Unknown or withheld values (15)

$.chaining not-recorded
$.confidence_note withheld
$.egress.terms not-recorded
$.extras.dead_end.retry_when process
$.free_tier.notes process
$.identity_cost.detail defaulted
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.interface_detail withheld
$.limits_note process
$.per_object_cap.notes not-recorded
$.per_object_cap.stated not-recorded
$.retention.notes process
$.retention.stated not-recorded
$.tooling not-recorded

Hosted inference APIs (as a compute class) Compute

deademailno cardRrequest-rate

groq.com/pricing/ · inference API · compute/inference-api-hosts · JSON · compare

Free tier	grouped row: an inference API answers tokens and is not a machine you can leave running unit tokens; metered on request-count; recorded so the census stops counting them as hosts
Price	none stated
Per-object cap	not applicable
Retention	no machine and no disk
Egress	not applicable
Identity	email an API key from an account
Limits	request-rate per-key and per-minute token limits set by each provider, as the earlier read recorded; this page does not restate them source no session, no disk, no SSH: the census counts this class by mistake
Binding	request-rate
Interface	ssh an HTTP completion API
Tooling	any OpenAI-compatible client
Composes with	belongs to the inference topic; it cannot run a shell or serve a file
Reliability	no reliability event recorded
Notes	A class-level dead end for this topic. The rows belong in docs/inference/, and this record exists so the compute census stops counting them.
Evidence rung	reported the site was read on 2026-10-22 and the extracted body is the marketing page rather than the pricing table, so this grouped record's rate limit statement cannot be re-confirmed here. The record is a grouping row and says so
Checked	2026-10-23 (live); stale on 2026-11-06

Evidence (2)

date	source	what it proved
2026-10-06	groq.com/pricing/	grouped because the census should stop counting them as hosts: an API you call is not a machine you leave running
2026-10-22	groq.com/pricing/	the Groq site was read on 2026-10-22 and carries its funding announcement and a console signup; the pricing table the record's URL names is not in the extracted body

More fields (8)

{
 "shape": null,
 "billing_model": "per-request",
 "session_cap": {
  "wall_clock_min": null,
  "idle_reap_s": null,
  "restart_required": false,
  "note": "an inference API answers tokens and holds no machine, which is why the record is a grouping row"
 },
 "keep_rate": null,
 "persistent_storage": null,
 "network_out": null,
 "free_tier_kind": "none",
 "census_rows": [
  "inference-apis"
 ]
}

Unknown or withheld values (3)

$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed