Pricing FAQ

In December 2017, we announced major price reductions along with the introduction of new features to Cloud Machine Learning Engine. Below is a summary of these price reductions and what they mean for Google Cloud customers.

FAQ

Why is Google Cloud lowering the prices for Cloud ML Engine?
We are determined to democratize machine learning (ML), and make it accessible and affordable. So we are updating our prices to make it easy for our customers to train and run any ML model and worry less about usage and costs. The improvements we are making in our infrastructure have made it possible for us to pass on the savings to you with these reduced prices.
How does this price reduction affect me?
There are two aspects of this price reduction. First, we have lowered the cost of training across the board for all types of computing resources and regions. This makes it very cost effective to train ML models in Google Cloud, regardless of the type of computing resource (GPUs and CPUs) required. Second, Google Cloud customers can run prediction models at low cost, regardless of the volume of predictions. Previously, we billed batch and online prediction by node hours and by number of predictions. We are now eliminating the charges based on the number of predictions, and reducing the node hour pricing. In summary, we have made it very affordable for Google Cloud customers to manage their ML lifecycle in the cloud from training to prediction.
When will the new prices be applied?
The new pricing for Cloud ML Engine is effective as of December 11, 2017.
What can I do to increase my quota?
Cloud ML Engine limits resource allocation and use, and enforces appropriate quotas on a per-project basis. If you need to increase your quota, please follow the guide to requesting a quota increase.

Price comparison

The following tables compare the old prices with the current pricing for training, batch prediction, and online prediction.

US

Training (US)
Predefined scale tiers - price per hour Old Current % price reduction
BASIC

$0.49

$0.2774

43%

STANDARD_1

$4.90

$2.9025

41%

PREMIUM_1

$36.75

$24.1683

34%

BASIC_GPU

$1.47

$1.2118

18%

Custom machine types - price per hour Old Current % price reduction
standard

$0.49

$0.2774

43%

large_model

$1.47

$0.6915

53%

complex_model_s

$0.98

$0.4141

58%

complex_model_m

$1.47

$0.8281

44%

complex_model_l

$2.94

$1.6562

44%

standard_gpu

$1.47

$1.2118

18%

complex_model_m_gpu

$5.88

$3.7376

36%

complex_model_l_gpu

$11.76

$7.4752

36%

standard_p100 (Beta)

---

$2.6864

---

complex_model_m_p100 (Beta)

---

$9.636

---

Batch prediction (US)
Old Current
$0.40 per node hour, plus $0.10 per thousand predictions (or $0.05 per thousand predictions for high volume) $0.09262 per node hour, with zero added cost per prediction.
Online prediction (US)
Old Current
$0.40 per node hour, plus $0.10 per thousand predictions (or $0.05 per thousand predictions for high volume) $0.3 per node hour, with zero added cost per prediction.

Europe

Training (Europe)
Predefined scale tiers - price per hour Old Current % price reduction
BASIC

$0.54

$0.3212

41%

STANDARD_1

$5.40

$3.3609

38%

PREMIUM_1

$40.50

$27.9794

31%

BASIC_GPU

$1.62

$1.3578

16%

Custom machine types - price per hour Old Current % price reduction
standard

$0.54

$0.3212

41%

large_model

$1.62

$0.8001

51%

complex_model_s

$1.08

$0.4795

56%

complex_model_m

$1.62

$0.9589

41%

complex_model_l

$3.24

$1.9179

41%

standard_gpu

$1.62

$1.3578

16%

complex_model_m_gpu

$6.48

$4.1464

36%

complex_model_l_gpu

$12.96

$8.2928

36%

standard_p100 (Beta)

---

$2.9784

---

complex_model_m_p100 (Beta)

---

$10.6288

---

Batch prediction (Europe)
Old Current
$0.44 per node hour, plus $0.11 per thousand predictions (or $0.05 per thousand predictions for high volume) $0.10744 per node hour, with zero added cost per prediction.
Online prediction (Europe)
Old Current
$0.44 per node hour, plus $0.11 per thousand predictions (or $0.05 per thousand predictions for high volume) $0.348 per node hour, with zero added cost per prediction.

Asia Pacific

Training (Asia Pacific)
Predefined scale tiers - price per hour Old Current % price reduction
BASIC

$0.54

$0.3212

41%

STANDARD_1

$5.40

$3.3609

38%

PREMIUM_1

$40.50

$27.9794

31%

BASIC_GPU

$1.62

$1.3578

16%

Custom machine types - price per hour Old Current % price reduction
standard

$0.54

$0.3212

41%

large_model

$1.62

$0.8001

51%

complex_model_s

$1.08

$0.4795

56%

complex_model_m

$1.62

$0.9589

41%

complex_model_l

$3.24

$1.9179

41%

standard_gpu

$1.62

$1.3578

16%

complex_model_m_gpu

$6.48

$4.1464

36%

complex_model_l_gpu

$12.96

$8.2928

36%

standard_p100 (Beta)

---

$2.9784

---

complex_model_m_p100 (Beta)

---

$10.6288

---

Batch prediction (Asia Pacific)
Old Current
$0.44 per node hour, plus $0.11 per thousand predictions (or $0.05 per thousand predictions for high volume) $0.10744 per node hour, with zero added cost per prediction.
Online prediction (Asia Pacific)
Old Current
$0.44 per node hour, plus $0.11 per thousand predictions (or $0.05 per thousand predictions for high volume) $0.348 per node hour, with zero added cost per prediction.

What's next

  • For more details of the pricing model and related resources, visit the Cloud Machine Learning Engine pricing page.
Was this page helpful? Let us know how we did:

Send feedback about...

Cloud ML Engine for TensorFlow