Machine learning as a service (MLaaS) platforms offer machine learning tools as a part of cloud computing services. If you are new to the MLaaS concept, check out our comprehensive beginner’s guide to machine learning as a service.
The key players in this market include Amazon, Microsoft, Google, and IBM. The offerings of these providers vary. So, if you are looking for the MLaaS platform that will be the best fit for your machine learning projects, make sure to go through the Comparing Machine Learning As A Service article with its detailed comparison of services offered by the key MLaaS providers.
Here, we want to provide you with another piece of the puzzle to help you make a final decision. Let’s have an overview of the pricing policies for the major MLaaS platforms to see what you can get for free and how much you need to pay to get access to more computing power and advanced capabilities.
Do you like this in-depth educational content on applied machine learning? Subscribe to our Enterprise AI mailing list to be alerted when we release new material.
Amazon Machine Learning
Free tier. Amazon offers its customers an opportunity to gain free hands-on experience with the AWS platform, products, and services. The free tier in machine learning includes the following offers, among others:
- 2 months’ free trial with Amazon SageMaker for building, training, and deploying ML models, including 250 hours per month of notebook usage, 50 hours per month for training, and 125 hours per month for hosting.
- 12 months’ access to Amazon Comprehend for training and managing NLP models, including 50K units of text (5M characters) for each API per month.
- 12 months’ access to Amazon Lex for voice and text chatbots, including 10K text requests per month and 5K speech requests per month.
- 12 months’ access to Amazon Rekognition for deep-learning-based image recognition (5K images per month included).
- 12 months’ access to Amazon Translate for high-quality neural machine translation (2M characters per month included).
- 2 months’ access to Amazon SageMaker Ground Truth for data labeling, with the first 500 objects per month labeled for free.
Paid services. Amazon has a pay-as-you-go approach for the pricing of its cloud services. So, with Amazon Machine Learning, you don’t pay any fixed amounts per month but only pay for the services you consume. Thus, the prices are mostly set per hour of each particular service. Note that the price varies based on your region and the computing capacity you use.
To understand the price level, it might be useful to go through a specific example.If, for a particular project, you use an ml.t2.medium Jupyter notebook for 105 hours, train the model four times for 30 minutes on an ml.m4.4xlarge and deploy to a ml.t2.medium for 10 minutes each time for the evaluation, you will spend $7.15 on model developing, training, and hosting. Next, if we assume that 3 GB of data is prepared into a notebook, 2 GB are pushed into Amazon S3, 1 GB of data is used for evaluation, and 1/10 of the input data is used for inference, you’ll need to pay an additional $0.08 for pulling training data into notebooks and moving data into Amazon S3, and $0.14 for data processing in and out of notebooks and hosting. So, the total amount for the whole workflow will be around $7.37. This doesn’t look like a huge amount, but if you are working with big data and deep neural networks that require more computing power, the price might increase significantly.
Azure Machine Learning
Free tier. Microsoft offers 25+ services that are always free and also provides the possibility to build, train, and deploy machine learning models for free for 12 months. During this trial period, you get a number of products for free, including:
- 750 hours of Linux or Windows Virtual Machines for computations.
- 64 GB × 2 of managed disk storage, 5 GB of Blob storage, and 5 GB of file storage.
- 250 GB for SQL databases and 5 GB for Azure Cosmos databases.
In addition, you get a $200 credit for exploring any Azure service for 30 days.
Paid services. When using Azure by Microsoft, you will also pay only for what you need – there are no upfront costs. Again, the prices vary greatly depending on the region as well as computing power and memory requirements. For example, to deploy ML models using Azure Kubernetes Service (AKS), you’ll need to pay from $0.05/hour up to $6+/hour.
Google Cloud Machine Learning
Free tier. Google offers a $300 free credit to try any Google Cloud Platform (GCP) product during the first 12 months. In addition, part of the products is “always free”, implying limited access to common GCP resources free of charge. In the machine learning and artificial intelligence part, free products include:
- Vision AI. 1K units per month for label detection, face detection, etc.
- Cloud Speech-to-Text. 60 minutes per month of speech-to-text transcription.
- Natural Language. 5K units per month of unstructured text processing.
Paid services. Google introduces price details for each of the GCP products. Several examples:
- Training jobs in all Americas regions are $0.49 per hour, per training unit.
- Machine translation costs $20 per million characters.
- Image analysis costs $1.5 per image for each feature (e.g., label detection, text detection, facial detection, etc.).
- Sentiment analysis is priced at $0.5-$1 per 1K queries depending on the number of queries.
- Content classification within the Natural Language API costs $0.1-$2 per 1K units depending on the number of units processed.
IBM Watson Machine Learning
Free tier. IBM allows its users to create a free Lite account that includes limited access to 40+ IBM Cloud services. These services include, among others:
- Free cluster with one worker node within Kubernetes Service to deploy apps in a native Kubernetes environment.
- 10K API calls per month for Watson Assistant to automate interactions with end users.
- 100 minutes per month of Watson Speech to Text service for converting human voice to written text.
- 10K per month of Watson Text to Speech service for synthesizing natural-sounding speech from text.
Paid services. When you upgrade from the Lite account to Pay-As-You-Go, you get access to 190+ IBM cloud services and a $200 credit valid for 30 days and applicable to any service. Out of 190 services, 15 are related to machine learning. Some of the examples are:
- IBM Watson Assistant for building and deploying virtual assistants, costing $0.0025 per message.
- IBM Watson Visual Recognition for object detection, costing $0.002 per use.
- IBM Watson Language Translator, with a price of $0.02 per 1K characters.
- IBM Natural Language Understanding for extracting metadata such as concepts, entities, and sentiment, with a price of $0.003 per 10K characters.
Additionally, in contrast to other MLaaS platforms, IBM has a subscription option, where you can get predictable billing and discounted rates in exchange for longer-term commitments.
Enjoy this article? Sign up for more updates on applied ML.
We’ll let you know when we release more technical education.