Azure Cognitive Services pricing

Use intelligence APIs to enable vision, language and search capabilities


For more options on the individual API offers, select an offer from the drop-down above. ​

QnA Maker, Speech Services and Custom Vision Service are currently only available as individual APIs.

Product Features Price
Computer Vision S1
up to 10 requests per second
Image type
0-1M transactions — $- per 1,000 transactions
1M-10M transactions — $- per 1,000 transactions
10M-100M transactions — $- per 1,000 transactions
100M+ transactions — $- per 1,000 transactions
Detect, objects
0-1M transactions — $- per 1,000 transactions
1M-10M transactions — $- per 1,000 transactions
10M-100M transactions — $- per 1,000 transactions
100M+ transactions — $- per 1,000 transactions
0-1M transactions — $- per 1,000 transactions
1M+ transactions — $- per 1,000 transactions
Content moderator S0
up to 10 requests per second
Moderate, Review 0-1M transactions - $- per 1,000 transactions
1M-5M transactions - $- per 1,000 transactions
5M-10M transactions - $- per 1,000 transactions
10M+ transactions - $- per 1,000 transactions
Face Standard
up to 10 requests per second
Face detection
Face verification
Face identification
Face grouping
Similar face search
0-1M transactions - $- per 1,000 transactions
1M-5M transactions - $- per 1,000 transactions
5M-100M transactions - $- per 1,000 transactions
100M+ transactions - $- per 1,000 transactions
Face storage $- per 1,000 faces per month

Computer Vision S1: Customers are charged per transaction not per API call. Learn more about what transactions are below.

* Products in Preview

+ Non-English languages are in Preview

** TPS only applies to web endpoint

Product Features Price
Language Understanding Standard
up to 50 requests per second
Text requests $- per 1,000 transactions
Speech requests $- per 1,000 transactions
Text Analytics Standard
up to 100 requests per second and 1,000 requests per minute
Sentiment analysis (and opinion mining)
Key phrase extraction
Language detection
Named entity recognition (not available in Container)
0-500,000 text records — $- per 1,000 text records
0.5M-2.5M text records — $- per 1,000 text records
2.5M-10.0M text records — $- per 1,000 text records
10M+ text records — $- per 1,000 text records
Translator S1 Text translation
Language detection
Bilingual dictionary
$- per million chars (Pay as you go)

Support and SLA

  • Free billing and subscription management support are included.
  • We guarantee that Cognitive Services running in the standard tier will be available at least 99.9 per cent of the time. No SLA is provided for the free trial. Read the SLA


Computer Vision API

  • Please refer to the documentation for more detailed descriptions of these Computer Vision features.

    • Adult – Detects adult/racy content to enable automated restriction in images.
    • Analyse – Calls multiple features at once. Specify which features you want to run and the API will run all of these together. Each feature included in “Analyse” will be counted as a separate transaction.
    • Celebrity – Recognises 200,000 celebrities from business, politics, sports and entertainment around the world.
    • Colour – Extracts colours from an image. The colours are analysed in three different contexts: foreground, background and whole. The colours are grouped into 12 dominant accent colours.
    • Face – Analyses human faces within an image.
    • GetThumbnail – Generates a high-quality thumbnail after an image is uploaded. It analyses the objects within the image, then crops the image to fit the requirements of the region of interest (ROI).
    • Image type – Indicates whether an image is black and white or colour, as well as use the same method to indicate whether an image is a line drawing or not. Indicates whether an image is clipart or not, and the quality.
    • OCR & Read – Both features apply optical character recognition (OCR) technology for detecting text in an image, which can be extracted for multiple purposes. The Read feature delivers highest accuracy for printed and handwritten text extraction. Please refer to Documentation for supported languages for OCR and Read.
    • Spatial analysis – Understand how people move through a physical space in near-real time.
    • Tag – Returns tags based on more than 2,000 recognisable objects, living beings, scenery and actions. In cases where tags may be ambiguous or not commonly known, the API response provides “hints” to clarify the meaning of the tag.
  • Each feature you select is counted as a transaction. There are a few special cases to note:

    1. Analyse allows you to select multiple features at once. For instance, an Analyse call specifying the Tag, Face and Adult features would count as three transactions.
    2. Read allows you to upload multipage PDF documents. Each page is counted as a feature. For instance, a 200-page document would count as 200 transactions.
    3. All GET calls to see the results of the async Read and Recognise text features are counted as transactions but are free of charge.
  • Each operation that you call (either individually or through “Analyse”) will be counted as a transaction. The total bill will be based on the number of transactions for each type of operation within a monthly billing period.

    As a specific example, let’s say you make the following calls in a certain monthly billing period:

    • 1,500,000 Analyse operations, each calling both Tag and Describe operations
    • 500,000 OCR operations
    • 4,000,000 Recognise text operations

    Your total bill will be constructed as follows:

    Operations Resource Calculations Subtotal
    1,500,000 Tag and 1,500,000 Face operations: S1 transactions First 1,000,000 transactions: $-/1000 * 1,000,000 = $-
    Remaining 2,000,000 transactions: $-/1000 * 2,000,000 = $-
    500,000 OCR operations: S2 transactions $-/1000 * 500,000 = $- $-
    1,500,000 Describe and 4,000,000 Recognise Text operations: S3 transactions $-/1000 * 5,500,000 = $- $-
    Total $- $-

Content moderator

  • When using the API, images need to have a minimum of 128 pixels and a maximum file size of 4 MB. Text can be at most 1024 characters long.

  • The text API will return an error code that informs that the text is longer than permitted. The image API will also return an error code that informs that the image does not meet the size requirements.

  • Human review tool is included in your subscription.

Face API

  • For operations that enable training at million-scale (available from 1 March 2018), a transaction is counted for every 1,000 images trained. “LargePersonGroup – Train” and “LargeFaceList – Train” belong to this category. Each operation in this category is rounded up to the nearest increment of 1,000 images. For example, a “LargeFaceList – Train” operation that includes 849,500 images will be counted as 850 transactions.

    For all other operations, each API call will be counted as a transaction.

    Please refer to the documentation for the complete list and detailed descriptions of operations.

  • Face Storage allows a subscription to store additional persisted faces when using person objects and face lists for identification, or for similarity matching with the Face API.

  • The stored faces are charged at $- per 1,000 faces, and this rate is prorated daily.

    For example, if your account used 10,000 persisted faces each day for the first half of the month and none the second half, you would be billed only for the 10,000 faces for the days stored. The calculation then is ($-/1,000) * (10,000*15+0*16)/31 = $-

    As a second example, if each day during the month you persist 1,000 faces for a few hours and then delete them each night, you would still be billed for 1,000 persisted faces each day.

  • The quota for the number of stored person groups is now 1 million, with up to 1 million persons per person group or face lists.


  • For text requests, a transaction is an API call with query length up to 500 characters.

    For speech requests, a transaction is an utterance with query length up to 15 seconds long.

  • No, the free tier only includes text requests with max. length of 500 characters.

  • Dispatch is a feature that enables processing two models/applications with one API call.

Text Analytics

  • The Text Analytics API can be purchased in units of the S0-S4 tier at a fixed price. Each unit of a tier comes with included quantities of API transactions. If the user exceeds the included quantities, overages are charged at the rate specified in the pricing table above. These overages are prorated and the service is billed on a monthly basis. The included quantities in a tier are reset each month. In the S tier, the service is billed for only the amount of Text Records submitted to the service.
  • Usage is throttled if the transaction limit is reached on the free tier. Customers cannot accrue overages on the free tier.
  • Any annotation to a document counts as a transaction. Batch scoring calls will also take into consideration the number of documents that need to be scored in that transaction. So for instance, if 1,000 documents are sent for sentiment analysis in a single API call, that will count for 1,000 transactions. If an API supports more than one annotation operation, that will also be considered. Let’s say an API call performs both sentiment analysis and key-phrase extraction on 1,000 documents, that will count for 2,000 transactions (2 annotations × 1,000 documents). The maximum size of a single document is 5,120 characters as measured by StringInfo.LengthInTextElements. If a document has more than 5,120 chars then we do not process that specific document and show invalid document as the error. If a request has multiple documents, then we process all documents which are less than 5,120 characters.
  • If the usage on the S0-S4 tier is exceeded, the account starts to accrue overages. These overages are billed on a monthly basis and are calculated at the rate specified for each tier.
  • You may upgrade to a higher tier at any time. The billing rate and included quantities corresponding to the higher tier will begin immediately.
  • A text record in the S tier contains up to 1,000 characters as measured by StringInfo.LengthInTextElements. If an input document into the text analytics API is more than 1,000 characters, it counts as one text record for each unit of 1,000 characters. For instance, if an input document sent to the API contains 7,500 characters, it would count as eight text records. If an input document sent to the API contains 500 characters, it would count as one text record. If two documents are submitted, one document of 500 characters and one document of 1,200 characters, then the service would be billed for three text records in total: one record for the 500 character document and two text records for the 1,200 character document.

Bing Custom Search

  • No, Bing Custom Search is curtailed to meet specific offerings of each Tier. For example, Tier S3 is meant for customers wanting to utilise only custom web results and custom image results for their selected set of domains. The S3 Tier also offers statistics capabilities. The API response will differ depending on the endpoint called within a specific tier. Each endpoint has its own response schema.

  • Here are the different endpoints:

    • Bing Custom Search (Custom Web Search endpoint)
    • Bing Custom Search with Image Search feature (Custom Image Search endpoint)
  • Tiers are priced based on the number of transactions. As an example, for Tier S3, the price per 1,000 transactions is $6 and for Tier S4, the price per 10,000 transactions is $4. At the end of the billing period, if 20,000 transactions are logged for the Tier S3 and 40,000 transactions are logged for the Tier S4, then the total will be $136 calculated as $6*(20,000/1,000) + $4*(40,000/10,000).

  • Statistics is offered as part of Tier S2 and Tier S3. Within 24 hours of doing a Bing Custom Search Web or Image API call using a Tier S2 or Tier S3 key, the charts in your authoring portal will start populating statistics. You can gather metrics such as call volume, top queries, API response code distribution and market distribution amongst many others.

  • The Custom Autosuggest feature is offered as a separate tier than Tier S4. Custom Autosuggest can be purchased separately and can be incorporated to get auto-suggestions for either S1, S2 or S3 Tiers. For example, if a customer wants to use S1 Tier and S4 tier, they will need to buy both the tiers separately and the billing will work as follows:

    If in S1 Tier, a customer has used a total of 3,000 transactions and in S4 tier, a customer has used a total of 40,000 transactions, the total billing will be $28 calculated as $4*(3,000/1,000) + $4*(40,000/10,000).


  • Standard Translation is an easy one-step process, in which you provide an input sentence and Azure’s pre-trained models translate them into a target language. By contrast, Custom Translation is a two-step process, in which first you train your own models using training data that you upload to the translation service. Once your model has been trained, you can perform the translation as you would with Standard Translation.

  • There are three billed activities that you can perform during Custom Translation:

    1. Model hosting: Hosting a model means that it is available to use for Custom Translation. You are charged a flat fee for every model that is hosted during a billing period. This is NOT pro-rated if the model is hosted for less than the full month.
    2. Training: Every time you train a model, you are charged a fee for every character in the training data. You are charged for characters in both the source and target languages of the training set, but there is a cap to how much you can be charged for any given training run, regardless of how many characters are in your training data. This cap applies to each training run, i.e. you would be charged if you were to rerun the same set.
    3. Translation: You are charged for every character of text translated by your Custom Translation model.
  • The S1-S4 tiers are designed to provide discounts for users who require high volumes of Standard Translation – not Custom – every month. While a discount is offered on the Standard Translation rate, S1-S4 tiers do not include a discount on Custom Translation. The C2-C4 tiers are intended to provide discounts to customers who regularly perform high-volumes of Custom Translation. While a discount is offered on the Custom Translation rate, C2-C4 tiers do not include a discount on Standard Translation.

  • In order to receive volume discounts on both Standard and Custom Translation, you are required to allocate both an S1-S4 instance AND a C2-C4 instance, directing Standard Translation to the S instance and Custom Translation to the C instance.


Estimate your monthly costs for Azure services

Review Azure pricing frequently asked questions

Learn more about Azure Cognitive Services

Review technical tutorials, videos and more resources

Added to estimate. Press 'v' to view on calculator

Talk to a sales specialist for a walk-through of Azure pricing. Understand pricing for your cloud solution.

Get free cloud services and $200 in credit to explore Azure for 30 days.