Microsoft Unveils New AI Models and Tools at Build 2025

During the Microsoft Build 2025 event held in Seattle from May 19 to 22, the spotlight was on AI agents, showcasing the company's commitment to enhancing developer tools and platforms. Central to the announcements was the Azure AI Foundry, previously known as AI Studio, which aims to provide developers with a comprehensive full-stack tool. This platform connects seamlessly with Visual Studio Code, GitHub, and Azure, streamlining the development process.

The Build event revealed an array of new models, functionalities, and interfaces designed to expand capabilities and improve interoperability. Among the notable innovations is a new Model Router that intelligently selects the most suitable model for a given prompt, enhancing user experience and efficiency.

New models introduced within Foundry include Grok 3 from xAI, Flux Pro 1.1 from Black Forest, and Sora from OpenAI, which is specifically designed for video applications. Microsoft is also previewing o4-mini, a tool for Reinforcement Fine-Tuning (RFT), although currently available only in select regions. Access to over 10,000 models from Hugging Face has also been made available, enriching the ecosystem for developers.

On the platform side, Microsoft is enhancing orchestration and infrastructure between agents and data. This includes the implementation of open standards such as Agent-2-Agent (A2A) and Model Context Protocol (MCP), which simplify communication with data artifacts on Google Cloud Platform (GCP) and Amazon Web Services (AWS).

Foundry will also facilitate the management of multi-data and agent workflows by integrating Microsoft's Semantic Kernel with AutoGen. A planned agent catalog will enable developers and architects to find code examples and configurations from various manufacturers, while local platforms can utilize a model runtime on Windows or Mac systems.

In addition to these advancements, Microsoft is prioritizing security and governance within its AI frameworks. New tools for observability, agent identity management, and trust assurance are being introduced. A dedicated Red Teaming Agent will scrutinize code for security vulnerabilities and oversee governance and bias considerations prior to deploying agents. This initiative is built on the open-source tool PyRIT, which tests model responses to potentially malicious queries. Additional tools will monitor AI systems for corrupted prompts and the inadvertent sharing of sensitive corporate information.

Furthermore, Microsoft is launching a fully managed Agent Service, which will handle infrastructure, orchestration, and data resources. This service can integrate with SharePoint, Microsoft Fabric, and other third-party systems, allowing for seamless connectivity to applications such as Office 365, Slack, and Twilio.

The future of AI agents is also being explored, with the introduction of an agent named Amelie, designed to create and deploy instant agents autonomously. Amelie can respond to queries and prompts by selecting appropriate models, training them quickly, and integrating the results into workflows. Microsoft is also engaged in an open-source project called Magnetic UI, which aims to develop an open human-machine interface.