Groq and Cerebras, from their docs
Both companies host an id spelled gpt-oss-120b. The prices and context windows below are not interchangeable: Groq’s id is openai/gpt-oss-120b, and Cerebras’s id is gpt-oss-120b. Nothing here ranks the two hosts.
| Groq | Cerebras | |
|---|---|---|
| gpt-oss-120b input / output per 1M | $0.15 / $0.60 | $0.35 / $0.75 |
| gpt-oss-120b context | 131,072 tokens | 65k free tier, 131k paid (model card) |
| gpt-oss-120b free or trial requests | Free Plan Limits payload: 30 RPM, 1K RPD, 8K TPM, 200K TPD | Rate-limits page: 5 RPM, 30K uncached TPM. Model card: 30 requests/min, 60k input tokens/min, 1M tokens/day |
| Smaller GPT-OSS id | openai/gpt-oss-20b at $0.075 / $0.30, 131,072 context | Not on the two Cerebras cards fetched |
| Account with no card | Free Plan Limits exist as a tab. The page does not say a card is required for that plan. | Free Trial FAQ: Playground and API stay inactive until a verified payment method is added. Credits are $5 for 30 days. No renewing free tier. |
| Data use, as stated | Services agreement: inputs and outputs are not used for training unless the customer grants permission. A zero-data-retention setting exists for eligible customers. | Privacy policy: inputs and outputs for inference services are not retained. |
Groq’s rate-limits page also says, above the table, that the limits shown are the base limits for the Developer plan. The selected tab on the fetched HTML was Free Plan Limits, and those cells match the freeRows payload, not the devRows payload (1K RPM and 250K TPM for openai/gpt-oss-120b). Cerebras’s two gpt-oss-120b limit sentences are both left in the table because the pages disagree.
Directory entries: Groq and Cerebras. Free-tier writeups: Groq and Cerebras.
Other host notes
- OpenRouter and Together, from their docs
- DeepInfra and Together listed prices
- Token price calculator
- Context windows
- LLM APIs category list
Sources
Last checked 2026-10-22.