Microsoft Azure AI is a portfolio, not one interchangeable service. Use a task-specific Foundry Tool for a defined job such as translation, document extraction, or speech transcription; Azure AI Search to retrieve relevant material; a Foundry model such as Azure OpenAI to generate or reason over content; Foundry Agent Service to connect a model with tools; and Azure Machine Learning when you need to build or train a bespoke model. The right starting point depends on the input and output your application needs.
What are Microsoft Azure AI services?
Microsoft groups prebuilt and customizable APIs and models as Foundry Tools. They support capabilities such as language processing, search, translation, speech, vision, document understanding, content safety, and decision-making. Separate parts of the broader portfolio provide foundation-model access, agent tooling, retrieval, and custom machine learning. These capabilities solve different parts of an application; choosing an AI service does not automatically supply every component an application needs.
Microsoft’s current documentation uses Microsoft Foundry as the unifying platform terminology, while older Azure AI and Cognitive Services references may use earlier names. Check the current service documentation for the exact name and product placement of the capability you intend to use. Microsoft’s Foundry Tools overview describes the prebuilt capabilities.
Which Azure AI service should you use?
Start with the result you need, then select the service family that produces it. These are starting points, not guarantees that every feature or model is available in every region or for every file type.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems#1 Best Overall
| What you need | Good starting point | Why it fits |
|---|---|---|
| Analyze text for sentiment, key phrases, entities, summaries, classification, language, question answering, or conversational intent | Azure Language in Foundry Tools | It provides targeted natural-language capabilities. For document search, Microsoft points to Azure AI Search; for translation, use Translator. Microsoft Learn: Azure Language. |
| Translate text or documents | Azure Translator in Foundry Tools | It supports real-time text translation, batch or single-file document translation, and custom translation for specialized terminology. Microsoft Learn: Translator. |
| Extract fields, tables, or structure from forms and documents | Azure Document Intelligence in Foundry Tools | It offers prebuilt document models as well as custom model options for extraction. Microsoft Learn: Document Intelligence. |
| Extract schema-defined fields from varied documents or media using natural-language descriptions | Azure Content Understanding in Foundry Tools | Consider it when a suitable prebuilt Document Intelligence model does not fit, or when a workflow needs confidence scores, grounding, or RAG-ready Markdown. Check the current documentation for supported inputs and availability. Microsoft Learn: Content Understanding. |
| Transcribe audio, synthesize speech, translate speech, or build speech interaction | Azure Speech in Foundry Tools | The service includes speech-to-text, text-to-speech, speech translation, and speaker recognition capabilities. Microsoft Learn: Speech. |
| Analyze image or video content | Azure Vision in Foundry Tools; consider Content Understanding for broader media extraction | Microsoft groups Vision and Content Understanding in its image and video processing guidance. Microsoft’s AI technology choices guide. |
| Search a collection of documents or retrieve relevant material for a conversational application | Azure AI Search | It indexes and retrieves relevant content. Microsoft includes it in retrieval-augmented generation guidance and directs document-search use cases to it. Microsoft’s AI technology choices guide. |
| Check harmful or unwanted text or image content | Content Safety in Foundry Control Plane | Microsoft describes it for checking user-generated and AI-generated content. Confirm current product placement and availability in the documentation. Microsoft’s Foundry Tools overview. |
| Generate, summarize, reason over, or understand content with a foundation model | Azure OpenAI in Foundry Models, or another suitable Foundry Model | Foundry provides managed model access and a broader model catalog. Select a specific model and confirm its availability in your intended region. Microsoft Learn: Microsoft Foundry. |
| Build an agent that uses a model plus tools or knowledge | Foundry Agent Service | It hosts agents connected to a model and, optionally, custom knowledge stores or APIs. Microsoft Learn: Foundry Agent Service. |
| Train a bespoke model or customize beyond what a prebuilt tool supports | Azure Machine Learning | It is the custom machine-learning path when prebuilt capabilities do not meet the requirement, and generally calls for more machine-learning expertise than using a prebuilt API. Microsoft’s AI technology choices guide. |
How to choose the right capability
- Define the output. Be specific: extracted document fields, translated text, a transcript, image labels, text sentiment, grounded answers over private content, or generated content. The required output usually narrows the service family faster than starting with a product name.
- Prefer a matching prebuilt tool. If a Foundry Tool already supports the job, it is usually a more direct starting point than building and operating a custom model. Microsoft notes that many projects can use prebuilt models or SaaS capabilities, with customization available for some services. Microsoft’s AI technology choices guide.
- Separate retrieval from generation. Azure AI Search indexes and retrieves relevant material; a language model generates or reasons over content. For answers grounded in a private document collection, assess retrieval quality and model behavior as distinct parts of the workflow. A model by itself should not be assumed to search your corpus. Microsoft’s AI technology choices guide.
- Choose custom machine learning for a reason. Use Azure Machine Learning when the prebuilt options cannot meet the requirement and a tailored model or training approach offers enough value to justify the additional data, expertise, operations, and governance work.
- Check deployment constraints before implementation. Verify the service and feature in your intended region, model availability, pricing and quota, API version, data handling, security controls, and retirement notices. Portfolio overviews do not establish those details for a particular deployment.
How Azure AI Search, Azure OpenAI, and Machine Learning differ
These products can work together, but they do different jobs. Search retrieves material; a foundation model generates or reasons over content; custom machine learning is for workloads that need a model or training approach beyond the prebuilt offerings.
| Capability | Primary role | Use it when |
|---|---|---|
| Azure AI Search | Indexing and retrieval | Your application needs to find relevant items in a document collection or supply retrieved material to a conversational workflow. |
| Azure OpenAI or another Foundry Model | Foundation-model access | Your application needs model-based generation, summarization, reasoning, or content understanding. |
| Azure Machine Learning | Custom model development and training | Prebuilt capabilities are insufficient and you need a bespoke model or training approach. |
A grounded-answer application may combine retrieval and generation rather than choosing one in place of the other. An agent can add tool or knowledge connections, but it does not eliminate the need to choose suitable models, retrieval, safety measures, and evaluation.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What to compare when two services seem to fit
Use the workload’s constraints to distinguish overlapping options, then check current service documentation for deployment-specific details.
Quick Recap
- Input and output: text, documents, audio, images, or video; and classification, extraction, translation, retrieval, or generation.
- Model approach: prebuilt capability, customization of a supported service, or training a bespoke model.
- Grounding: whether the answer must be based on a private or maintained collection of material.
- Data fit: supported languages and file formats for the exact feature you plan to use.
- Deployment fit: regional availability, data residency, volume, latency, cost, and quota.
- Operations and controls: API lifecycle, identity, network isolation, safety, and monitoring requirements.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
