Foundry Models Overview

Explore Microsoft Foundry’s comprehensive model catalog including Azure OpenAI, Foundry Direct, partner, and community models.

Foundry Models Overview

Microsoft Foundry Models is your destination for discovering, evaluating, and deploying AI models for building agents, copilots, and AI applications.

Model Catalog

Access 1900+ models from Microsoft, OpenAI, and leading AI companies:

Models Sold Directly by Azure

Hosted and sold by Microsoft with direct support, SLAs, and deep Azure integration

Partner & Community Models

Specialized models from Anthropic, Meta, Cohere, Hugging Face, and more

Model Categories

Foundation Models

• GPT-4o: Multimodal reasoning and generation
• GPT-4: Advanced language understanding
• Claude: Long-context processing (200K tokens)
• Llama 3: Open-source alternative

Reasoning Models

• Multi-step problem solving
• Mathematical reasoning
• Code generation and analysis

Small Language Models (SLMs)

• Phi-3 family: Efficient edge deployment
• Lower latency and cost
• Specialized tasks

Multimodal Models

• Process text and images
• Vision and language combined
• Examples: GPT-4o, Claude 3

Deployment Options

Serverless API

• Pay-per-token billing
• Instant scaling
• No capacity management
• Best for: Development, variable workloads

Provisioned Throughput

• Reserved model capacity
• Predictable performance and cost
• Fungible across compatible models
• Best for: Production, high-volume

Managed Compute

• Deploy to dedicated VMs
• Open-source model support
• Full control
• Best for: Custom deployments

Model Selection Guide

1. Define Requirements

   Task, inputs, outputs, latency needs

2. Check Capabilities

   Input types, context length, features

3. Evaluate Performance

   Benchmarks, testing with your use case

4. Assess Cost

   Token costs, volume estimates

5. Verify Availability

   Regional availability, lifecycle status

Key Features

• Tool Calling: GPT-4o, GPT-4, Claude 3
• JSON Mode: Structured output guarantee
• Vision: Image processing (GPT-4o, Claude)
• Streaming: Incremental response delivery
• Fine-Tuning: Customize with your data

See Model Deployment and Region Support for details.