Cerebras Cloud inference (shared)

Fields

id

services/svc-cerebras-cloud

Provider

cerebras

Endpoint

https://api.cerebras.ai/v1

Input $/Mtok

0.35

Output $/Mtok

0.75

Price basis

USD per 1M TOKENS, representative model gpt-oss-120b, first-party on the Cerebras model page, re-verified 2026-10-17. The page prints Input $0.35 / Output $0.75 for both the Free tier and Pay as you go. Same numbers as the 2026-10-14 seed, which had to take them from third-party price trackers because the public pricing page did not extract a per-model rate table that day. qwen-3.8-27b is the other Shared Inference model; no first-party per-1M rate extracted for it on 2026-10-17, so it is not the representative price. Dedicated Inference remains sales-gated.

Context window

131,072

Output tok/s

3,000

Measured by

vendor

Regions

notes

2026-10-17: price is now first-party. inference-docs.cerebras.ai/models/openai-oss.md prints gpt-oss-120b at $0.35 in / $0.75 out on Free and on Pay as you go, speed ~3000 tok/s, context 65k free / 131k paid. The dollar figures are unchanged from the 2026-10-14 seed; what changed is the source. The old basis said the public pricing page did not extract a per-model table and the $0.35/$0.75 came from third-party trackers at lower confidence. The model catalog overview fetched the same day still lists exactly two Shared Inference models (gpt-oss-120b at ~3000 tok/s, qwen-3.8-27b at ~1850 tok/s) and still says everything else is Dedicated Inference. measured_tps stays the vendor ~3000, not a third-party median: Artificial Analysis still measures Cerebras in the 2000+ tok/s class on this model, which makes the vendor figure plausible rather than marketing-inflated, but it is still vendor. qwen-3.8-27b is named in notes only — there is no data/models slug for that id, and models_served is a cross-reference.

Notes

2026-10-17: price is now first-party. inference-docs.cerebras.ai/models/openai-oss.md prints gpt-oss-120b at $0.35 in / $0.75 out on Free and on Pay as you go, speed ~3000 tok/s, context 65k free / 131k paid. The dollar figures are unchanged from the 2026-10-14 seed; what changed is the source. The old basis said the public pricing page did not extract a per-model table and the $0.35/$0.75 came from third-party trackers at lower confidence. The model catalog overview fetched the same day still lists exactly two Shared Inference models (gpt-oss-120b at ~3000 tok/s, qwen-3.8-27b at ~1850 tok/s) and still says everything else is Dedicated Inference. measured_tps stays the vendor ~3000, not a third-party median: Artificial Analysis still measures Cerebras in the 2000+ tok/s class on this model, which makes the vendor figure plausible rather than marketing-inflated, but it is still vendor. qwen-3.8-27b is named in notes only — there is no data/models slug for that id, and models_served is a cross-reference.

Cross-references

Models (1)

ModelsGPT-OSS 120B