Pricing for Azure Content Understanding in Foundry Tools

This article explains the Azure Content Understanding in Foundry Tools pricing model. Learn what you're charged for and how to estimate costs for your workload.

For specific pricing rates, see Azure Content Understanding Pricing.

Unless otherwise noted, this article references GA behavior with API version 2025-11-01. Sections that describe workflow-family resolution and agentic mode switch to preview behavior with API version 2026-06-01-preview.

Understand the two types of charges

Azure Content Understanding pricing is based on two main usage categories:

1. Content extraction charges

Content extraction transforms unstructured input (documents, audio, video) into structured, searchable text and content. This output includes optical character recognition (OCR) for documents, speech-to-text for audio/video, and layout detection. You pay per input unit processed:

• Documents: Per 1,000 pages
• Audio and video: Per minute

2. Generative feature charges

When you use AI-powered features that call large language models (LLMs), you incur two types of charges:

• Contextualization charges: Prepares context, generates confidence scores, provides source grounding, and formats output. For details, see Contextualization tokens.
• Generative model charges: Token-based costs from Microsoft Foundry model deployments (LLMs for generation, embeddings for training examples). Content Understanding uses the Foundry model deployment you provide for all generative AI-related calls. You don't see any LLM or embedding token usage billing in Content Understanding. That usage appears on your Foundry model deployment. For details, see Generative model charges.

Generative features include: Field extraction, figure analysis, segmentation, categorization, training.

Note

The "agentic" workflow (preview) resolves to the agentic.* workflow family and uses the advanced contextualization rate. It can also consume more model tokens than a nonagentic workflow. Test with representative documents when you estimate costs. For how labeled data and agentic mode affect the resolved workflow, see Custom analyzers.

Cost components

Your total cost includes each usage category that applies to your analyzer: content extraction, contextualization tokens, completion model input and output tokens, and embedding model tokens. If you use only content extraction without generative capabilities, you pay only for content extraction.

How to estimate your costs

1. Test with representative files

Run a small test analysis with your actual files and schema. To see usage for GA and preview operations, check the usage object in the Analyzers API response:

{
  "usage": {
      "documentPagesMinimal": 0, // Asynchronous: Digital file extraction.
      "documentPagesBasic": 0, // Asynchronous: Read extraction.
      "documentPagesStandard": 2, // Asynchronous: Layout extraction.
      "documentPagesMinimalInline": 0, // Preview: Synchronous digital file extraction.
      "documentPagesBasicInline": 0, // Preview: Synchronous Read extraction.
      "documentPagesStandardInline": 0, // Preview: Synchronous Layout extraction.
      "contextualizationTokens": 2000,
      "advancedContextualizationTokens": 0, // Preview: Advanced contextualization.
      "tokens": {
         "gpt-5.2-input": 10400,
         "gpt-5.2-output": 360
    }
  }
}

In the 2026-06-01-preview API, synchronous operations report content extraction usage in the documentPages*Inline properties. Advanced contextualization usage is reported in advancedContextualizationTokens. These property names identify usage in the API response; they aren't the names of published pricing meters.

2. Get current rates

Use the pricing pages for Azure Content Understanding and Foundry Models to get the current rates for your region, deployment type, and selected models.

3. Calculate the estimate

For each applicable usage category, multiply the measured usage by its current rate. Include content extraction and contextualization charges from Content Understanding, completion model input and output token charges, and embedding token charges when your analyzer uses embeddings. Add the resulting charges to estimate the total cost for your expected volume.

Rates vary by region, deployment type, and model, so this article doesn't provide numerical calculations. For current model options, see Supported generative models.

Detailed cost components

Content extraction

Content extraction is the essential first step for transforming unstructured input—whether it's a document, audio, or video—into a standardized, reusable format. This foundational processing is required for all generative features and can be used standalone.

Content extraction pricing by modality:

• Documents: Three tiered meters (minimal, basic, or standard) based on processing complexity
• Audio: Speech-to-text transcription (single standard meter, priced per minute)
• Video: Frame extraction, shot detection, and speech-to-text transcription (single standard meter, priced per minute)
• Images: No content extraction available

Document content extraction meters

For documents, you're charged for the type of processing Content Understanding performs. Content Understanding charges based on the actual work performed on each page, not the analyzer you select.

Minimal meter: Applies to digital documents (DOCX, XLSX, PPTX, HTML, TXT, MSG, EML) where no OCR or layout processing is needed. This meter is the lowest-cost option for digital-native documents. You're charged the minimal rate regardless of which analyzer you use—even if you call a layout analyzer on a digital document, you're only charged for the minimal processing performed.

Basic meter: Applies when Content Understanding performs OCR processing to extract text from image-based documents (scanned PDFs, images, TIFFs) without layout analysis.

Standard meter: Applies when Content Understanding performs layout analysis, including table recognition and structural element detection from image-based documents (scanned PDFs, images, TIFFs).

The following table shows which meter applies based on your file type and analysis level:

File Type	Read (Basic)	Layout (Standard)
Image-based (PDF, PNG, TIFF, JPG, and other image-based formats)	Basic meter	Standard meter
Digital formats (DOCX, XLSX, HTML, TXT, and other digital formats)	Minimal meter	Minimal meter

Tip

The meter charged depends on the processing Content Understanding actually performs, not which analyzer you choose. Digital documents always use the minimal meter because they don't require OCR or layout processing.

Generative capabilities

The generative capabilities of Content Understanding use generative AI models to enhance the quality of the output. In the latest API version 2025-11-01, you can choose a generative model based on your use case.

When you use any generative capabilities, Content Understanding uses the Foundry models deployment you provide. The token usage for the completion or embeddings models is on that deployment.

Contextualization tokens

Contextualization is the processing layer in Content Understanding that preprocesses user content, prepares data and context for generative models, and post-processes model output into the final structured results. The type of data processing, data preparation, and context processing applied determines the type and amount of contextualization tokens charged.

Standard contextualization applies general-purpose extraction capabilities to your content and schema.

Advanced contextualization applies specialized technology to improve quality, simplify implementation, and solve more complex content understanding problems.

With the 2026-06-01-preview API, the resolved config.workflow value makes the applicable rate explicit:

• Workflow values that start with standard use the standard contextualization rate.
• All other workflow families, including advanced and agentic, use the advanced contextualization rate.

Customers set "default" or "agentic" when creating an analyzer. The service returns a versioned workflow family value, such as standard.2026-06-01-preview, advanced.2026-06-01-preview, or agentic.2026-06-01-preview. For the complete resolution rules, see workflow.

When you're charged: Whenever you use generative capabilities (field extraction, figure analysis, segmentation, categorization, training).

Pricing: Fixed rate per content unit.

Contextualization tokens are calculated per unit of content:

Units	Contextualization tokens	Effective standard contextualization price per unit	Effective advanced contextualization price per unit
Per page	1,000 contextualization tokens	$1 per 1,000 pages	$3 per 1,000 pages
Per image	1,000 contextualization tokens	$1 per 1,000 images	$3 per 1,000 images
Per hour of audio	100,000 contextualization tokens	$0.10 per hour	Not applicable
Per hour of video	1,000,000 contextualization tokens	$1 per hour	Not applicable

Assuming a standard contextualization rate of $1.00 per 1 million contextualization tokens and an advanced contextualization rate of $3.00 per 1 million contextualization tokens.

Prebuilt analyzers

The following table summarizes the prebuilt analyzers and their applicable content extraction and context processing billing meters. The field extraction meter is charged based on actual usage of the generative model.

Analyzer name	Content extraction meter	Contextualization meter
Content extraction analyzers
prebuilt-layout	Document Standard	None
prebuilt-read	Document Basic	None
prebuilt-digitalParse	Document Minimal	None
Base analyzers
prebuilt-audio	Audio	None
prebuilt-document	Document Minimal/ Basic/ Standard (depending on file type and analyzer configuration)	None
prebuilt-image	None	None
prebuilt-video	Video	None
RAG analyzers
prebuilt-documentSearch	Document Minimal/ Basic/ Standard (depending on file type and analyzer configuration)	Standard Contextualization
prebuilt-imageSearch	None	Standard Contextualization
prebuilt-audioSearch	Audio	Standard Contextualization
prebuilt-videoSearch	Video	Standard Contextualization
Domain-specific analyzers
prebuilt-invoice	Document Standard	Standard Contextualization
prebuilt-receipt	Document Standard	Standard Contextualization
prebuilt-receipt.generic	Document Basic	Standard Contextualization
prebuilt-receipt.hotel	Document Standard	Standard Contextualization
prebuilt-creditCard	Document Basic	Standard Contextualization
prebuilt-creditMemo	Document Standard	Standard Contextualization
prebuilt-check.us	Document Standard	Standard Contextualization
prebuilt-bankStatement.us	Document Standard	Standard Contextualization
prebuilt-idDocument	Document Basic	Standard Contextualization
prebuilt-idDocument.generic	Document Basic	Standard Contextualization
prebuilt-idDocument.passport	Document Basic	Standard Contextualization
prebuilt-healthInsuranceCard.us	Document Standard	Standard Contextualization
prebuilt-tax.us	Document Standard	Standard/Advanced Contextualization
prebuilt-tax.us.1040	Document Standard	Standard Contextualization
prebuilt-tax.us.1040Senior	Document Standard	Standard Contextualization
prebuilt-tax.us.1040Schedule1	Document Standard	Standard Contextualization
prebuilt-tax.us.1040Schedule2	Document Standard	Standard Contextualization
prebuilt-tax.us.1040Schedule3	Document Standard	Standard Contextualization
prebuilt-tax.us.1040Schedule8812	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleA	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleB	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleC	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleD	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleE	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleEIC	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleF	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleH	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleJ	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleR	Document Standard	Standard Contextualization
prebuilt-tax.us.1040ScheduleSE	Document Standard	Standard Contextualization
prebuilt-tax.us.1099Combo	Document Standard	Standard Contextualization
prebuilt-tax.us.1099A	Document Standard	Standard Contextualization
prebuilt-tax.us.1099B	Document Standard	Standard Contextualization
prebuilt-tax.us.1099C	Document Standard	Standard Contextualization
prebuilt-tax.us.1099CAP	Document Standard	Standard Contextualization
prebuilt-tax.us.1099DA	Document Standard	Standard Contextualization
prebuilt-tax.us.1099DIV	Document Standard	Standard Contextualization
prebuilt-tax.us.1099G	Document Standard	Standard Contextualization
prebuilt-tax.us.1099H	Document Standard	Standard Contextualization
prebuilt-tax.us.1099INT	Document Standard	Standard Contextualization
prebuilt-tax.us.1099K	Document Standard	Standard Contextualization
prebuilt-tax.us.1099LS	Document Standard	Standard Contextualization
prebuilt-tax.us.1099LTC	Document Standard	Standard Contextualization
prebuilt-tax.us.1099MISC	Document Standard	Standard Contextualization
prebuilt-tax.us.1099NEC	Document Standard	Standard Contextualization
prebuilt-tax.us.1099OID	Document Standard	Standard Contextualization
prebuilt-tax.us.1099PATR	Document Standard	Standard Contextualization
prebuilt-tax.us.1099Q	Document Standard	Standard Contextualization
prebuilt-tax.us.1099QA	Document Standard	Standard Contextualization
prebuilt-tax.us.1099R	Document Standard	Standard Contextualization
prebuilt-tax.us.1099S	Document Standard	Standard Contextualization
prebuilt-tax.us.1099SA	Document Standard	Standard Contextualization
prebuilt-tax.us.1099SB	Document Standard	Standard Contextualization
prebuilt-tax.us.1099SSA	Document Standard	Standard Contextualization
prebuilt-tax.us.1098	Document Standard	Standard Contextualization
prebuilt-tax.us.1098E	Document Standard	Standard Contextualization
prebuilt-tax.us.1098T	Document Standard	Standard Contextualization
prebuilt-tax.us.1095A	Document Standard	Advanced Contextualization
prebuilt-tax.us.1095C	Document Standard	Advanced Contextualization
prebuilt-tax.us.w2	Document Standard	Standard Contextualization
prebuilt-tax.us.w4	Document Standard	Standard Contextualization
prebuilt-tax.us.1041ScheduleK1	Document Standard	Advanced Contextualization
prebuilt-tax.us.1120SScheduleK1	Document Standard	Advanced Contextualization
prebuilt-tax.us.1065ScheduleK1	Document Standard	Advanced Contextualization
prebuilt-tax.us.8865ScheduleK1	Document Standard	Advanced Contextualization
prebuilt-tax.us.mn.m1	Document Standard	Advanced Contextualization
prebuilt-mortgage.us	Document Standard	Standard Contextualization
prebuilt-mortgage.us.1003	Document Standard	Standard Contextualization
prebuilt-mortgage.us.1004	Document Standard	Standard Contextualization
prebuilt-mortgage.us.1005	Document Standard	Standard Contextualization
prebuilt-mortgage.us.1008	Document Standard	Standard Contextualization
prebuilt-mortgage.us.closingDisclosure	Document Standard	Standard Contextualization
prebuilt-contract	Document Standard	Standard Contextualization
prebuilt-marriageCertificate.us	Document Standard	Standard Contextualization
prebuilt-procurement	Document Standard	Standard Contextualization
prebuilt-purchaseOrder	Document Standard	Standard Contextualization
prebuilt-payStub.us	Document Standard	Standard Contextualization
prebuilt-utilityBill	Document Standard	Standard Contextualization
Utility analyzers
prebuilt-documentFieldSchema	Document Standard	Standard Contextualization
prebuilt-documentFields	Document Standard	Standard Contextualization

¹Extracting content from text-based files by using supported analyzers is billed with the Document Minimal meter.

Custom analyzers

When you build custom analyzers and provide labeled training data, the system applies the advanced contextualization meter. Otherwise, it applies the standard contextualization meter.

Custom analyzers created with API version 2025-11-01 or earlier continue to use the standard contextualization rate. If you retrieve one of these analyzers with the 2026-06-01-preview API, the response shows standard.2025-11-01 in config.workflow. The 2025-11-01 API response is unchanged.

When you create a custom analyzer with the 2026-06-01-preview API:

• A custom analyzer without labeled data resolves to standard.2026-06-01-preview and uses the standard rate.
• A custom analyzer with labeled data resolves to advanced.2026-06-01-preview and uses the advanced rate.
• A custom analyzer created with workflow set to "agentic" resolves to agentic.2026-06-01-preview and uses the advanced rate.

Generative model charges (LLM)

Token-based charges from Foundry models that power the actual field extraction, analysis, and other generative capabilities.

Input tokens include:

• Extracted text and transcripts
• Image tokens (for visual analysis)
• Your schema definitions
• System prompts
• Labeled training examples with the 2025-11-01 API

Output tokens include:

• Field values and structured data
• Confidence scores and source grounding
• Analysis results and descriptions

Cost optimization: Compare the current rates for supported models and deployment types, and choose an option that meets your quality, latency, data residency, and cost requirements.

Embeddings charges

Token-based charges for embedding models used when training custom analyzers with labeled examples to improve accuracy.

• When charged: Only when using the training feature with labeled data
• Models: For current options, see Supported generative models.
• Typical usage: The entire document is embedded. Usage can vary depending on the density of text, but about 1,500 tokens per page is a useful initial estimate.

Generative feature details

Each generative feature has different cost implications.

Field extraction

Generates structured key-value pairs based on your schema definition. Examples include invoice sender/receiver, line items, or video ad elements like tagline and product appearance.

Cost impact: Charges scale with schema complexity and content size.

Figure analysis

Creates descriptive text for images, charts, and diagrams to make visual content searchable in RAG workflows.

Cost impact: LLM tokens per image analyzed - both input tokens for image interpretation and output tokens for descriptions. Usage scales with the size and number of images contained in the document.

Segmentation

Divides documents or videos into logical sections for targeted processing and improved efficiency.

Cost impact: Output token costs for each segment created. Optionally, you can chain analyzers for further analysis on each segment. Chaining uses more content extraction and generative usage, equivalent to running the chained analyzers independently.

Categorization

Assigns labels to documents or segments for classification and intelligent routing to specialized analyzers.

Cost impact: LLM and contextualization costs for classification. Routing to another analyzer adds its respective charges.

Training

Builds custom analyzers using labeled examples for domain-specific accuracy improvements.

Cost impact: Both API versions use embedding tokens when you add labeled data. With the 2025-11-01 API, retrieved training examples also add LLM input tokens during analysis. With the 2026-06-01-preview API, the built analyzer doesn't retrieve the labeled documents at analysis time.

Frequently asked questions

When am I charged for LLM usage?

You're charged for LLM tokens only when you provide the analyzer with a Foundry deployment and use a generative capability in Content Understanding. Analyzers that only perform content extraction, such as prebuilt-read, prebuilt-layout, or custom analyzers without generative capabilities, don't incur LLM charges.

How do I know which content extraction meter applies to my documents?

The actual processing performed determines the meter, not the analyzer you choose:

• Minimal: Digital documents (DOCX, XLSX, HTML, TXT, etc.) always use minimal, regardless of analyzer
• Basic: Image-based documents with OCR-only processing (Read analyzer)
• Standard: Image-based documents with layout analysis (Layout analyzer)

For more information about meters, see Document content extraction meters.

Am I charged twice for Foundry model usage?

No. Content Understanding uses the LLM deployments linked for all LLM and embedding calls. You're billed on those deployments. You pay Content Understanding for content extraction and contextualization, and Foundry for the generative model tokens (input/output tokens and embeddings).

How does model choice affect cost?

Model token rates vary by model and deployment type. Compare current rates and test representative files before choosing a model. Content extraction and contextualization charges don't depend on the completion model you select.

What increases token usage?

Several features can increase token consumption:

• Source grounding and confidence scores add processing context.
• Extractive mode adds instructions and output details.
• Training examples add retrieved examples to the model context by using the 2025-11-01 API.
• Segmentation and categorization can require additional model calls.

Am I charged if my request fails?

Content Understanding doesn't charge for content extraction or contextualization when a request fails with an error (such as a 400 error). If a Foundry completion model call succeeded before the failure, you're charged for that Foundry model usage based on Foundry's billing policies.

Cost optimization tips

• Compare supported models and deployment types to balance cost, quality, latency, and data residency requirements.
• Start with mini models - Mini models offer substantial savings for most extraction tasks
• Use global deployments when data residency and compliance allows
• Enable advanced features selectively - Only use source grounding and confidence scores when needed
• Test representative files before scaling to understand actual token consumption
• Monitor usage regularly through the Azure portal to identify optimization opportunities

More pricing examples

Here are detailed examples showing how pricing works across different scenarios:

Example 1: Document processing for RAG workflows

Scenario: You need to extract content from documents for a Retrieval-Augmented Generation (RAG) solution. You use prebuilt-documentSearch to extract text, layout, and figure descriptions.

Input:

• 10 pages
• Model: GPT-5.2 global deployment
• Region: East US

Pricing breakdown:

1. Content extraction: 10 pages

   • Cost: (10 / 1,000) × $5.00 = $0.05

2. Figure analysis:

   Assuming two figures per page. It costs about 1000 input and 200 output tokens per figure.

   • Input tokens: 20 figures × 1,000 tokens/image = 20,000 tokens
   • Cost: (20,000 / 1,000,000) × $2.00 = $0.04
   • Output tokens: 20 figures × 200 tokens/figure = 4,000 tokens
   • Cost: (4,000 / 1,000,000) × $8.00 = $0.032

3. Contextualization: 10 pages × 1,000 tokens/page = 10,000 tokens

   • Cost: (10,000 / 1,000,000) × $1.00 = $0.01

Total estimated cost: $0.05 + $0.04 + $0.032 + $0.01 = $0.132

Note

These prices are for illustration purposes only and aren't intended to represent the actual cost. Check Azure Content Understanding Pricing and Azure OpenAI Pricing for current rates

Example 2: Processing invoices with field extraction

Scenario: You're automating invoice processing using prebuilt-invoice to extract structured data (invoice number, date, vendor, total, line items).

Input:

• 10 pages
• Model: GPT-5.2 global deployment
• Features: Extractive mode + source estimation + confidence scores
• Region: East US

Pricing breakdown:

1. Content extraction: 10 pages

   • Cost: (10 / 1,000) × $5.00 = $0.05

2. Field extraction: with source estimation + confidence enabled, the token usage is ~2x more per page:

   • Base input tokens: 10 pages × 5,200 tokens/page = 52,000 tokens
   • Cost: (52,000 / 1,000,000) × $0.40 = $0.0208
   • Base output tokens: 10 pages × 180 tokens/page = 1,800 tokens
   • Cost: (1,800 / 1,000,000) × $1.60 = $0.0029

3. Contextualization: 10 pages × 1,000 tokens/page = 10,000 tokens

   • Cost: (10,000 / 1,000,000) × $1.00 = $0.01

Total estimated cost: $0.05 + $0.0208 + $0.0029 + $0.01 = $0.0837

Note

Model token rates vary. Use the current rate for your selected model when you estimate field extraction costs.

Note

These prices are for illustration purposes only and aren't intended to represent the actual cost. Check Azure Content Understanding Pricing and Azure OpenAI Pricing for current rates

Example 3: Analyzing video content with segment-level field extraction

Scenario: You're extracting a structured representation of video content for a RAG application. To extract structured data per segment of video, you can use the prebuilt-videoSearch. Segments are short clips of 15-30 seconds on average, resulting in numerous output segments with a single summary field per segment.

Input:

• 60 minutes (1 hour) of video
• Model: GPT-5.2 global deployment
• Region: East US

Assumptions:

• Input tokens: 7,500 tokens per minute (based on sampled frames, transcription, schema prompts, and metaprompts)
• Output tokens: 900 tokens per minute (assuming 10-20 short structured fields per segment with auto segmentation)
• Contextualization: 1,000,000 tokens per hour of video

Pricing breakdown:

1. Content extraction: 60 minutes

   • Cost: 60 minutes × $1/hour = $1.00

2. Field extraction:

   • Input tokens: 60 minutes × 7,500 tokens/minute = 450,000 tokens
   • Cost: (450,000 / 1,000,000) × $2.00 = $0.90
   • Output tokens: 60 minutes × 900 tokens/minute = 54,000 tokens
   • Cost: (54,000 / 1,000,000) × $8.00 = $0.43

3. Contextualization: 1,000,000 tokens per hour

   • Cost: (1,000,000 / 1,000,000) × $1.00 = $1.00

Total estimated cost: $1.00 + $0.90 + $0.43 + $1.00 = $3.33

Note

Actual cost varies based on the specifics of your input and output. This transparent, usage-based billing model ensures you only pay for what you use.

Note

These prices are for illustration purposes only and aren't intended to represent the actual cost. Check Azure Content Understanding Pricing and Azure OpenAI Pricing for current rates

Example 4: Processing audio call center recordings

Scenario: You're analyzing call center recordings using prebuilt-callCenter to generate transcripts, speaker diarization, sentiment analysis, and summaries.

Input:

• 60 minutes of audio
• Model: GPT-5.2 global deployment
• Region: East US

Pricing breakdown:

1. Content extraction: 60 minutes

   • Cost: 60 minutes × $0.36/minute = $0.36

2. Field extraction:

   • Input tokens: 60 minutes × 604 tokens/minute = 36,240 tokens
   • Cost: (36,240 / 1,000,000) × $0.40 = $0.01
   • Output tokens: 60 minutes × 19 tokens/minute = 1,140 tokens
   • Cost: (1,140 / 1,000,000) × $1.60 = $0.00

3. Contextualization: 60 minutes × 1,667 tokens/minute = 100,020 tokens

   • Cost: (100,020 / 1,000,000) × $1.00 = $0.10

Total estimated cost: $0.36 + $0.01 + $0.00 + $0.10 = $0.47

Note

These prices are for illustration purposes only and aren't intended to represent the actual cost. Check Azure Content Understanding Pricing and Azure OpenAI Pricing for current rates

Example 5: Processing images with captions

Scenario: You're generating descriptive captions for product images using prebuilt-imageSearch.

Input:

• 1,000 images
• Model: GPT-5.2 global deployment
• Region: East US

Pricing breakdown:

1. Content extraction: No charge for images

   • Cost: $0.00

2. Field extraction:

   • Input tokens: 1,000 images × 1,043 tokens/image = 1,043,000 tokens
   • Cost: (1,043,000 / 1,000,000) × $2.00 = $2.09
   • Output tokens: 1,000 images × 170 tokens/image = 170,000 tokens
   • Cost: (170,000 / 1,000,000) × $8.00 = $1.36

3. Contextualization: 1,000 images × 1,000 tokens/image = 1,000,000 tokens

   • Cost: (1,000,000 / 1,000,000) × $1.00 = $1.00

Total estimated cost: $0.00 + $2.09 + $1.36 + $1.00 = $4.45

Usage categories by workload

Use the following table to identify the usage categories to include in your estimate. The exact token usage depends on your content, schema, analyzer configuration, and selected model.

Workload	Usage categories to review
Document extraction and RAG	Document content extraction, contextualization, completion input and output tokens, and embedding tokens when applicable
Structured field extraction	Content extraction, contextualization, and completion input and output tokens
Video analysis	Video content extraction, contextualization, and completion input and output tokens
Audio analysis	Audio content extraction, contextualization, and completion input and output tokens
Image analysis	Contextualization and completion input and output tokens

Next steps

• Azure Content Understanding Pricing page
• Content Understanding quickstart
• Best practices for Content Understanding
• Prebuilt analyzers