Model APIs - 9 entries
Cerebras API rate limits
Cerebras's official docs, read on October 17, 2026, set rate limits for Cerebras Inference across Free Trial, Developer, Private Preview, Pro and Max. For gpt-oss-120b, Free Trial is listed at 5 RPM, 30K Uncached TPM and 90K Total TPM, and Developer (Pay as You Go) at 1K RPM, 1M Uncached TPM and 3M Total TPM.
Last verified from inference-docs.cerebras.ai, cerebras.ai. What the docs leave out
- Tiers
- 5
- Tables
- 6
- Entries
- 9
- Verified
- Oct 17, 2026
Limits by model, endpoint and plan
Values are copied exactly as the docs show them, units included, in the docs' own order. A dash means the docs list no value for that row.
gpt-oss-120b
2 tiers
| Tier | RPM |
|---|---|
| Free Trial | 5 |
| |
| Developer (Pay as You Go) | 1K |
| |
- Free Trial: Free Trial: "New accounts receive $5 in free credits after adding a verified payment method. These credits expire 30 days after they're granted." No permanently free tier. Limits apply at the organization level and vary by model; any metric can trigger rate limiting. Dual-bucket model: uncached TPM is the primary limit, Total TPM (uncached + cached) defaults to 3x uncached. Token-bucket replenishment (continuous refill, not fixed reset). Estimated tokens include max_completion_tokens before the request runs. Exceeding returns 429 Too Many Requests indicating which bucket (uncached or total) was exceeded. Docs say exact org values are in Cloud Console > Limits and may differ from the table.
- Developer (Pay as You Go): Docs worked example: "With a 1M uncached TPM limit, your total limit is 3M TPM. At a 70% cache hit rate, you can effectively process up to 3M total tokens per minute - 1M uncached plus up to 2M cached." Org Admins can set lower per-project, per-model limits; project quota cannot exceed the org-level quota. Pricing page labels Developer tier "For development, evaluation, and experimentation; not intended for production use."
qwen-3.8-27b
2 tiers
| Tier | RPM |
|---|---|
| Free Trial | 5 |
| |
| Developer (Pay as You Go) | 300 |
| |
- Free Trial: Footnote verbatim: "Image limits: 2 per request, 10 MiB total request payload". Same Free Trial terms as gpt-oss-120b ($5 credit, expires 30 days after granted).
- Developer (Pay as You Go): Total TPM 750K is explicitly marked temporary; standard is 450K (3x uncached).
Batch API (/v1/chat/completions)
Private Preview
| Metric | Private Preview |
|---|---|
| Requests per batch | 50,000 requests maximum per batch |
| Minimum requests per file | Minimum 10 requests |
| File size | 200 MB maximum file size |
| Per request line | 1 MB maximum per request line |
| Concurrency | 10 concurrent active batches |
| Completion window | 24h ("Batch requests are guaranteed to complete in 24 hours.") |
| Result retention | Results retained for 7 days after completion |
| Rate limits | Configurable rate limits separate from real-time API |
- All requests in a file must use the same model; UTF-8 with LF line endings.
service_tier parameter (priority, default, auto, flex) on dedicated endpoints
Private Preview (Dedicated only)
| Metric | Private Preview (Dedicated only) |
|---|---|
| priority vs default | "priority and default rate limits are the same" |
| flex | "flex rate limits are tracked independently and are several multiples of default rate limits" |
| queue_threshold header (flex/auto) | Valid range: 50-20000 milliseconds |
- "Service tiers are not available with Shared Inference." During preview all service tiers are billed equally.
Projects (per-project, per-model quotas)
| Metric | Limit |
|---|---|
| Org-level quota | Hard limit for the entire organization (e.g., 1M TPM on a given model) |
| Project-level quota | Per-project limit; cannot exceed the org-level quota |
- By default projects have no explicit quotas and inherit org-level limits. All keys in a project share that project's rate limits; every request must satisfy both levels.
Cerebras Code (GLM 4.7)
2 tiers
| Tier | Tokens per day |
|---|---|
| Pro ($50) | Send up to 24million tokens per day |
| Max ($200) | Send up to 120 million tokens/day |
- Pro ($50): Marketing page copy: "enough for 3-4 hours of uninterrupted vibe coding". Plan shown as sold out; GLM 4.7 is not listed in the current Shared Inference model catalog in the docs, so treat as a legacy/unavailable plan. Free Trial on the same page: "GLM 4.7 access with limited tokens and requests" (no numbers).
- Max ($200): Shown as "sold out" on 2026-10-17. No RPM/TPM published for Code plans.
How to raise your limits
gpt-oss-120b and qwen-3.8-27b
Purchase credits (Pay as You Go) from the Billing tab in the Cloud Console (https://cloud.cerebras.ai). "Your first purchase moves you to the Developer tier with significantly higher rate limits and no hourly or daily token caps."
gpt-oss-120b and qwen-3.8-27b
For Enterprise limits or reserved capacity, contact Cerebras (https://www.cerebras.ai/contact) or your account representative.
Batch API (/v1/chat/completions) and service_tier parameter (priority, default, auto, flex) on dedicated endpoints
Private Preview: contact Cerebras or your account representative for access.
Projects (per-project, per-model quotas)
Only Org Admins can set project rate limits: All Projects view > Projects tab > Project Settings > set per-model limits > Save.
Cerebras Code (GLM 4.7)
Upgrade to Max ($200). Page shows both plans as "sold out" as of 2026-10-17.
What the docs leave out
- Exact per-organization limits are only shown in the logged-in Cloud Console (cloud.cerebras.ai > Limits); docs say console values reflect plan and project-level limits and can differ from the public table.
- Enterprise numeric limits are not published (provisioned per organization).
- No published RPH or RPD values for any tier, although the docs list RPH/RPD as measured metrics.
- Developer tier TPH/TPD: none (explicitly not applied).
- No published concurrency limit for real-time (non-batch) Shared Inference requests.
- Flex service tier exact multiples not published ("several multiples of default").
- Batch API separate rate limit values not published ("configurable").
- Cerebras Code Free Trial token/request numbers not published, and Cerebras Code Pro/Max RPM/TPM not published; both paid plans shown as sold out.
- Dedicated Inference model limits not published publicly.
FAQ
What are the Cerebras API rate limits?
Cerebras's official docs, read on October 17, 2026, set rate limits for Cerebras Inference across Free Trial, Developer, Private Preview, Pro and Max. For gpt-oss-120b, Free Trial is listed at 5 RPM, 30K Uncached TPM and 90K Total TPM, and Developer (Pay as You Go) at 1K RPM, 1M Uncached TPM and 3M Total TPM.
How do I raise my Cerebras Inference rate limits?
Purchase credits (Pay as You Go) from the Billing tab in the Cloud Console (https://cloud.cerebras.ai). "Your first purchase moves you to the Developer tier with significantly higher rate limits and no hourly or daily token caps." For Enterprise limits or reserved capacity, contact Cerebras (https://www.cerebras.ai/contact) or your account representative. Private Preview: contact Cerebras or your account representative for access.
Where do these Cerebras Inference numbers come from?
Every value is copied verbatim from Cerebras's official documentation at inference-docs.cerebras.ai and cerebras.ai, read on October 17, 2026. This page lists all 9 entries in 6 tables, each linked to the page it came from.
Related reading
Sources
Last verified . Every value is copied verbatim from the official pages below, read on that date.
Newsletter
Know when limits change
One email per week. New model releases, pricing and limit changes, and what they mean for what you build.