# Six jobs. Six specialized models.

> Azure AI Foundry pairs each business problem — search, images, video, claims, calls — with the model actually built for it.

- **Format:** short video, 8 steps, ~25 seconds
- **Topic:** Azure AI Foundry specialized models (embeddings, image generation, video generation, image analysis, speech) tied to real business problems
- **Author:** Suthahar Jegatheesan (MSDEVBUILD)
- **Category:** Azure · Azure
- **Published:** 2026-09-17
- **Tags:** azure, azureai, azureaifoundry, azureopenai, generativeai, computervision, speechtotext, texttospeech, artificialintelligence, microsoft, cloudcomputing, msdevbuild
- **Canonical URL:** https://blog.msdevbuild.com/shorts/azure-ai-foundry-specialized-models/

---
## What you'll learn

- Which Azure AI Foundry model fits which business problem — embeddings, image and video generation, vision, and speech.
- Why semantic search beats keyword search for real customer questions.
- How a multimodal model and speech models turn a photo or a phone call directly into structured, actionable data.

## Understand it one step at a time

### 1. Can one model handle six different jobs?

No. Search, images, video, claims and calls each need a different kind of model.

### 2. What does Azure AI Foundry actually offer?

Embeddings, image and video generation, vision, speech to text and text to speech — one catalog.

### 3. How do you search when the words don't match?

Ada and Cohere embeddings match by meaning, so "wet shoes" still finds the refund policy.

### 4. Need 30 product shots by tomorrow?

GPT-image-1 renders the full set from one prompt — no design queue required.

### 5. Can you test an ad before the shoot?

Sora 2 renders a 15-second teaser from one text prompt — no crew, no location, no wait.

### 6. One damage photo. What happens next?

GPT-4.1 reads the photo and drafts damage, severity and cost before anyone opens the ticket.

### 7. How do you handle support calls at scale?

GPT-4o-transcribe writes the call down; GPT-4o-tts speaks the confirmation back.

### 8. So who's actually using this?

Search, marketing, claims and support — one Foundry catalog behind all four.

---

## The takeaway

**Stop forcing one model to do six jobs**

Azure AI Foundry's catalog: embeddings for search, GPT-image-1 for visuals, Sora 2 for video, GPT-4.1 for vision, GPT-4o-transcribe and GPT-4o-tts for voice — one platform, one set of Azure controls.