Cerebras pricing

verified 2026-09-02

Wafer-scale hardware serving open models at extreme token throughput — speed is the headline, price stays competitive.

Official price list ↗Developer console ↗Price a workload

data residency & retention

from Cerebras's own policies · verified 2026-09-02 · some facts not published

Headquarters

🇺🇸 United States (US)

Where inference runs

Cerebras discloses US inference datacenters (Santa Clara, Stockton, Dallas, Minneapolis, Oklahoma City) and markets inference as running entirely on US-based infrastructure; first European capacity is promised by end of 2026, not live today.

Region pinning

Enterprise plans only

The self-serve API has a single endpoint with no region selection; in-region residency claims apply to enterprise dedicated and on-prem deployments only.

Prompt retention

Not retained

Privacy policy states inputs and outputs for inference services are not retained ('never stored, logged, or reused'); no abuse-log carve-out or retention window is disclosed.

Trains on API data

Terms state the content license does not grant Cerebras the right to use service content for training or fine-tuning models.

Zero data retention

Available

Zero data retention is the stated platform default, not a separate toggle or paid tier.

current models · $ per 1M tokens

Model	Input /1M	Cached in	Output /1M	Context
GPT OSS 120B GPT-OSS	$0.35	—	$0.75	131K
Gemma 4 31B Gemma	$0.99	—	$1.49	131K

Cerebras totals · 1M in + 1M out

GPT OSS 120B

$1.10

Gemma 4 31B

$2.48

Cost of 1M input + 1M output tokens. Bar length is square-root scaled so cheap models stay visible next to premium ones.