{"id":39946,"date":"2025-04-28T14:15:00","date_gmt":"2025-04-28T21:15:00","guid":{"rendered":"https:\/\/azure.microsoft.com\/en-us\/blog\/?p=39946"},"modified":"2025-12-26T12:51:41","modified_gmt":"2025-12-26T20:51:41","slug":"azure-ai-foundry-your-gps-for-the-changing-ai-landscape","status":"publish","type":"post","link":"https:\/\/azure.microsoft.com\/en-us\/blog\/azure-ai-foundry-your-gps-for-the-changing-ai-landscape\/","title":{"rendered":"Azure AI Foundry: Your GPS for the changing AI landscape"},"content":{"rendered":"\n

I spent most of the last decade in the Midwest (hello, Chicagoland!<\/em>) but now live in the Seattle area and work at the Microsoft campus in Redmond. Before moving here, I\u2019d visited the Redmond campus dozens of times as a Microsoft partner and customer, but had only basic airport-campus-hotel-restaurant navigation skills. Relocating a family of four meant a lot of exploring to find my way to grocery stores (what do you mean there\u2019s no Mariano\u2019s in Seattle?<\/em>), schools, doctors, dentists, etc. I couldn\u2019t have done it without GPS: Just plug in the destination and let it guide us, turn by turn, on the quickest, safest route, avoiding traffic and roadwork along the way.<\/p>\n\n\n\n

The hallmark of good tech is taking something complex and stressful (ever been lost in an unfamiliar city?<\/em>) and making it simple. Looking over the list of new AI capabilities we announced in just the past month, it\u2019s easy to see why turn-by-turn guidance is needed in this rapidly expanding AI landscape. Azure AI Foundry<\/a> is like GPS for your AI ambitions: it simplifies navigating the insane pace of innovation in the AI landscape so you see where you are today, evaluate what\u2019s new, and plan, execute, and manage your growing AI portfolio while optimizing costs. All with the confidence that it always has the latest AI thing.<\/p>\n\n\n\n

This wayfinding provides dexterity to help your teams stay nimble as AI presents new strategic opportunities and efficiencies. Here are just a few new capabilities in Azure AI Foundry<\/a> this month that show the value of managing your AI portfolio on a single do-it-all platform. <\/p>\n\n\n\n

\n
Try Azure AI Foundry today ><\/a><\/div>\n<\/div>\n\n\n\n

Models: More is more <\/h2>\n\n\n\n

As I write this there are 1,917 models in the Azure AI Foundry model catalog. Choosing one\u2014or three, since many AI apps now use multiple models\u2014is easier than you might think when facing an abundance of choice. On Azure AI Foundry, your team can quickly evaluate models on quality, cost, and throughput, or filter by industry, inference tasks like translation or text generation, or other key features to quickly get a short list of models to consider. New tools like the Evaluation API<\/a> in Azure OpenAI Service give developers the ability to run automated comparisons between OpenAI models, quantify quality improvements, and accelerate the iteration cycle. Together with the Stored Completions API<\/a>, the Evaluation API provides a programmatic solution for refining and evaluating AI models with accuracy and control. <\/p>\n\n\n\n

That\u2019s important when you consider we announced eight new Azure OpenAI Service models in one week\u2014the new GPT-4.1 model series<\/a>, three advanced audio models<\/a>, and the o3 and o4-mini reasoning models<\/a>. You can fine-tune GPT-4.1 and GPT-4.1-mini<\/a> to adapt them to your unique business requirements. And last week we announced GPT-image-1<\/a>, the latest and most advanced image generation model, as well as the latest GPT-4o-realtime-preview and GPT-4o-mini-realtime-preview API updates including WebRTC support, semantic voice activity detection and advanced input noise reduction and more.<\/p>\n\n\n\n

Check out how Costar Group uses Azure OpenAI Service models to transform their customers\u2019 experience of finding the right home.<\/p>\n\n\n\n

\n
See Homes.com transformed with Azure OpenAI<\/a><\/div>\n<\/div>\n\n\n\n

That much newness in one week might seem like an anxiety accelerator\u2014especially if your team just built an app using an earlier model\u2014but Azure AI Foundry was designed for the progressive, iterative flow of AI development. It\u2019s easy for teams to quickly evaluate new model pros and cons, drop the model in the playground for some hands-on testing, determine if it\u2019s better, faster, cheaper, etc., and deploy it if desired. Simple. And we intend to enable OpenAI models in Azure AI Foundry the same day as OpenAI announces them. They are backward-compatible and can be deployed with a single click or by adding a new model with a few simple lines of code.<\/p>\n\n\n\n

Azure OpenAI accounts for 35 of the 1,917 models in our catalog, but is not alone in introducing new models. Also new in our model catalog this month:<\/p>\n\n\n\n

The first models in the Meta Llama 4 herd<\/a><\/u><\/strong> arrived, including Llama 4 Scout<\/strong>, which specializes in condensing or analyzing extensive information, like analyzing all documents in an enterprise SharePoint library to answer a specific query, or reading a multi-thousand-page technical manual to provide troubleshooting advice. And Llama 4 Maverick<\/strong>, which specializes in image and text understanding with support for 12 languages for precise image understanding and creative writing.<\/p>\n\n\n\n

An incredible amount of knowledge sits in file cabinets around the world, invisible to AI. Mistral OCR<\/a>, a state-of-the-art Optical Character Recognition (OCR) model, can read and understand the structure and nuance of hard copy documents. It can extract equations, images, tables, and multilingual text from scanned contracts, handwritten forms, research papers, regulatory filings, and more at scale, making all that knowledge accessible for AI to reason over and your team to learn from and act on. <\/p>\n\n\n\n

Optimized for the enterprise, Cohere\u2019s Command A and Embed 4<\/a> models form a powerhouse for advanced apps like extracting insights from thousands of financial filings, earnings call transcripts, and market research. <\/p>\n\n\n\n

DeepSeek got the world\u2019s attention with the release of its R1 model earlier this year. It\u2019s a powerful and capable, low-cost model with some safety challenges requiring careful management. The Microsoft AI team developed its own DeepSeek R1 variant, MAI-DS-R1<\/a>, that outperforms DeepSeek R1 on responses to blocked topics and reducing harmful content while maintaining the same reasoning capabilities. You can find MAI-DS-R1 in Foundry today.<\/p>\n\n\n\n

Go to market faster with Azure AI Services<\/h2>\n\n\n\n