Foundation models that exceed benchmark performance across image, video, and text.
Phi
Small language models for building generative AI applications with better latency and lower costs.
Meta
Pre-trained, open language models ranging from 7 billion to 70 billion parameters.
Mistral AI
Accelerate AI innovation and achieve state-of-the-art reasoning performance.
Cohere
A leading large language model for retrieval-augmented generation capabilities.
DeepSeek
DeepSeek is a Chinese artificial intelligence company that trains models at a significantly lower cost. DeepSeek R1 is now available on Azure AI Foundry and GitHub.
NVIDIA NIM Microservices
NVIDIA NIM is a set of easy-to-use microservices designed to accelerate the deployment of generative AI across enterprises.
Hugging Face
Thousands of models spanning categories from text generation to image analysis.
Stability AI
Deliver exceptional text-to-image generation with superior quality and prompt adherence.
Nixtla
Pre-trained, generative AI transformer models for time-series analysis.
AI21
Foundation chat completion models for enterprise that accelerate the use of generative AI for production.
NTT Data
A high-performance, lightweight Japanese and English SLM with fine-tuning for secure hybrid deployment.
Core42, a G42 company
Leading Arabic language model JAIS accelerates the growth of a vibrant Arabic language AI ecosystem.
“We’re using Azure to build more and better AI models for RapidRead to help support the high demand on veterinary radiologists. Additionally, the benefit of scale helps us boost the accuracy of our AI models and expand our operations.”
Jerry Martin, VP of Research & Development, Mars
“When it came to AI models, the Azure AI platform was miles ahead for enterprise-segment offerings compared to alternatives.”
Lenin Gali, Chief Digital and Business Officer, Atomicwork
Foundry Models are a hub for discovering foundation models. The catalog includes some of the most popular large language and vision foundation models curated by Microsoft, OpenAI, DeepSeek, xAI, Hugging Face, Meta, Mistral AI, Cohere, Deci, Stability AI, Nixtla, and NVIDIA. These models are packaged for out-of-the-box use and are optimized for use in Azure AI Foundry.
MaaS or Serverless API is a deployment type that allows developers to access and use a variety of models hosted on Azure without having to provision GPUs or manage back-end operations. MaaS offers inference APIs and hosted fine-tuning for models such as Meta Llama2, Meta Llama 3, Mistral Large, and others.
MaaS charges are based on the number of tokens used for inference and the amount of data used for fine-tuning. The pricing varies depending on the model and the region. To see the price for individual models, search for the model in Azure Marketplace.