EU (multi-region) US (multi-region) Las Vegas (us-west4) Los Angeles (us-west2) Montréal (northamerica-northeast1) Northern Virginia (us-east4) Oregon (us-west1) Salt Lake City (us-west3) São Paulo (southamerica-east1) South Carolina (us-east1) Belgium (europe-west1) Finland (europe-north1) Frankfurt (europe-west3) London (europe-west2) Netherlands (europe-west4) Zürich (europe-west6) Hong Kong (asia-east2) Jakarta (asia-southeast2) Mumbai (asia-south1) Osaka (asia-northeast2) Seoul (asia-northeast3) Singapore (asia-southeast1) Sydney (australia-southeast1) Taiwan (asia-east1) Tokyo (asia-northeast1)
Operation Pricing Details
Logistic regression model creation 1 The first 10 GB of data processed by CREATE MODEL statements per month is free under the BigQuery free tier.
Linear regression model creation 1
K-means clustering model creation 1
Time series model creation 1, 2
AutoML Tables model creation 1 The total cost of a job is the sum of the following two costs:
  • BigQuery data processing cost (for preprocessing)
  • Passthrough AI Platform cost (for training)
The cost is converted to BigQuery ML bytes processed, and two billing records with different labels are reported for the job:
  • {key:, value:bqml_analytic} for BigQuery costs
  • {key:, value:cloud_service_name} for external AI Platform service costs
DNN model creation 1
Boosted tree model creation
Matrix factorization model creation Not supported Matrix factorization is only available to flat-rate customers or customers with reservations. On-demand customers are encouraged to use flex slots to use matrix factorization.
Evaluation, inspection, and prediction (all model types) Charges from evaluation, inspection, and prediction queries are included in the 1 TB of data per month under the BigQuery analysis free tier.
If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.

1 The CREATE MODEL statement stops at 50 iterations for iterative models.

2 For time series models, when auto-arima is enabled for automatic hyper-parameter tuning, multiple candidate models are fitted and evaluated during the training phase. In this case, the number of bytes processed by the input SELECT statement is multiplied by the number of candidate models, which can be controlled by the AUTO_ARIMA_MAX_ORDER training option.