Comparison of major services
Services
Feature Comparison
| Feature | Description | Azure AI Foundry | Amazon Bedrock | Google Vertex AI Studio |
|---|---|---|---|---|
| Model Catalog/Garden | Central listing of foundation models you can browse and deploy | ✅ | ✅ | ✅ |
| Playgrounds | Interactive UIs to test prompts or model inferences quickly | ✅ | ✅ | |
| Code/IDE | Strong support for applying the services in code (libraries, templates, containers, tutorials) | ✅ | ❓ | |
| Prompt flow | A streamlined tool for LLM application development, simplifying prototyping, experimentation, iteration, and deployment. | ✅ | ✅ (“Model chaining” via console/API) |
❓ |
| Tracing | Execution logs, lineage, or step-by-step run details for prompt-based apps | ✅ | ❓ | ❓ |
| Evaluation | Automated or semi-automated model performance checks | ✅ | ✅ | |
| Database | Built-in or highly streamlined database integration | ✅ | ❓ | ✅ |
| Web apps | Quick deployment to a web endpoint or basic UI for model demos | ✅ (Azure App Service integration)) |
❓ | ❓ |
| Fine-tuning | Update a model's task capabilities by adding new data via the weights it was initially trained with. This adapts the model to specific tasks or domains (e.g., legal text, marketing language, or enterprise jargon) without training it entirely from scratch. It refines the model’s weights so it can generate more relevant outputs based on your specialized data. | ✅ | ✅ | ✅ (Vertex AI “Model Tuning”) |
| Vectorstore / Embedding Index | Built-in or integrated way to store, query, and manage embeddings for similarity search (documents, images, etc.). | Uses Azure Cognitive Search or Azure Vector DB in preview | Amazon Kendra, OpenSearch Vector Engine | Vertex Matching Engine or partner solutions like Pinecone |
| Knowledge Bases | Query models against internal documents or knowledge stores (RAG) | ✅ (Azure AI Search) |
✅ (Vertex AI Vector Search) |
|
| Multi-modal | Support for images, text, audio, or other data types | ✅ | ✅ | |
| Data Automation | ✅ | |||
| Marketplace Deploy | 1-click or streamlined deployment of solutions to a marketplace | ❓ | ✅ | ❓ |
| Prompt Routers | Enables you to use a combination of foundation models for your generative AI applications to achieve better performance at lower cost and latency than any single model. | ❓ | ✅ | ❓ |
| Imported Models | Importing your model files from a blob/S3 bucket or by importing an AI/ML notebook model. | ❓ | ✅ | ✅ |
| Custom Models | Customize a model with Fine-tuning, Distillation or Continued Pre-training. | ❓ | ✅ | ✅ |
| Providers | Microsoft, OpenAI, Mistral, Meta, Stability, Core42, Nixtla, DeepSeek, Phi, Cohere, Hugging Face | Amazon, Anthropic, Cohere, Luma, Meta, Mistral, Stability | Google, SalesForce, Meta, Stability, Mistral, Anthropic, Hugging Face | |
| Pipelines | Visual or code-based MLOps workflows as a series of steps to train or deploy an ML model (e.g., inputs, outputs, logic) | ✅ | ||
| Model Monitoring | ✅ | |||
| Notebooks | ✅ | |||
| Feature Store | A centralized repository for creating, managing, and serving machine learning features. | ✅ | ||
| Agent Builder | Create AI agents and applications using natural language or a code-first approach. | ✅ | ||
| Vision | Build, deploy, and manage computer vision applications with a fully managed, end-to-end application development environment | ✅ | ||
| Translation | Translate text from one language to another |