Amazon SageMaker · Pricing Plans
Amazon Sagemaker Plans Pricing
Amazon SageMaker AI is priced per ML instance-hour across notebook, training, batch transform, real-time inference, and serverless inference workloads, with rates varying by instance family. SageMaker Savings Plans offer up to 64% off on a 1- or 3-year commitment. New accounts get a 2-month Free Tier including 250 hours of ml.t3.medium Studio notebooks and 50 hours of m4.xlarge or m5.xlarge training. Prices listed are us-east-1 examples.
3 Plans
API Commons Plans
View Source
AmazonMachine LearningAINotebooksInference
Plans
SageMaker Free Tier (2 months)
freemium
2-month introductory tier for new accounts.
Studio notebooks (ml.t3.medium) (instance-hour · month)
0.00 USD
Training (ml.m4.xlarge or ml.m5.xlarge) (instance-hour · month)
0.00 USD
- First 2 months of new account
- Includes Studio notebooks, training, real-time inference allowances
SageMaker On-Demand
usage-based
Pay-per-instance-hour across all SageMaker workloads. No upfront commitment.
Notebook instance (ml.c5.xlarge example) (instance-hour · usage)
0.204 USD
Training (ml.m5.xlarge example) (instance-hour · usage)
0.23 USD
Training (ml.m4.4xlarge example) (instance-hour · usage)
0.96 USD
GPU notebook / training (ml.g4dn.xlarge) (instance-hour · usage)
0.7364 USD
Real-time inference (ml.c5.xlarge example) (instance-hour · usage)
0.204 USD
Serverless inference (compute) (gb-second · usage)
see vendor pricing calculator USD
Serverless inference (data processing) (gb · usage)
see vendor pricing calculator USD
Other ML instance types (instance-hour · usage)
see vendor pricing calculator USD
- Instance-hour metering
- Distinct rates for notebook, training, batch transform, real-time, serverless inference
- GPU and Inferentia/Trainium families available
SageMaker Savings Plans
subscription
1- or 3-year commitment in $/hour for SageMaker compute in exchange for up to 64% off on-demand.
Savings Plans commitment (USD_per_hour_committed · year)
see vendor pricing calculator USD
- Up to 64% discount vs on-demand
- 1-year or 3-year terms
- Applies to Studio notebooks, training, inference