Key dev updates:<\/strong> Agent Service hits GA with Model Context Protocol (MCP) support\u2014integrate any backend with standardized JSON-RPC calls, no more custom OpenAPI specs. New models include o3-pro for enterprise reasoning, Sora for video generation, and DeepSeek-R1 for open-source reasoning. Platform changes:<\/strong> New unified Azure AI Foundry resource type with Projects for self-service development, simplified API version management with Never build alone, join the community!<\/strong> Connect with over 25,000 developers through Discord and GitHub Discussions, attend live events hosted by Microsoft teams, complete free open-source courses on AI fundamentals, and explore code samples to kickstart your AI applications. The community offers everything from live AMA sessions on Model Context Protocol to weekly .NET + AI office hours, plus comprehensive learning paths for AI Agents, MCP, and Generative AI development.<\/p>\n o3-pro<\/strong> delivers the highest reasoning performance<\/strong> by using more compute to provide deeper, more consistent answers across complex enterprise tasks. Features multimodal input support, structured outputs, tool integration, and enterprise-ready capabilities\u2014designed for organizations tackling their most challenging reasoning problems with precision and depth.<\/p>\n Sora is now available in public preview<\/strong> in Azure AI Foundry, bringing enterprise-grade video generation capabilities. Create up to 20-second videos in 1080p resolution with landscape, portrait, or square aspect ratios from text prompts. Currently supports text-to-video with image-to-video coming soon\u2014perfect for marketing campaigns, educational content, and creative workflows.<\/p>\n The new codex-mini<\/strong> model joins Azure OpenAI as a lightweight, efficient option for code generation, completion, and programming assistance\u2014ideal for embedded scenarios where performance and cost matter.<\/p>\n Grok 3<\/strong> and Grok 3 Mini<\/strong> from xAI are now available in Azure AI Foundry. These models excel at real-time conversational AI and general-purpose reasoning tasks with competitive performance and pricing.<\/p>\n DeepSeek-R1-0528<\/strong> brings the latest evolution in reasoning-optimized models with improved depth of reasoning and inferencing capabilities<\/strong>. This open-source model approaches leading models like OpenAI o3 and Gemini 2.5 Pro with enhanced chain-of-thought prompting, reinforcement learning fine-tuning, and broader multilingual understanding\u2014ideal for intelligent agents, copilots, and research applications.<\/p>\n Foundry Agent Service becomes first-class MCP clients<\/strong> with today’s preview. Bring any remote MCP server\u2014self-hosted or SaaS\u2014and Azure AI Foundry automatically imports capabilities, keeps them updated, and routes calls through enterprise security. Think “USB-C for AI integrations”\u2014the open JSON-RPC protocol lets you publish tools and resources once, then connect to any compliant agent runtime. No more hand-rolling Azure Functions or managing OpenAPI specs for every backend.<\/p>\n Seamlessly integrate agents with Microsoft SharePoint<\/strong> to chat with private documents securely. Powered by the Microsoft 365 Copilot API, agents can connect to SharePoint sites with intelligent indexing, query processing, and content chunking. Features identity passthrough (On-Behalf-Of) authentication for proper access control, ensuring users only access documents they have permission to view\u2014maintaining enterprise-grade security while grounding AI responses in up-to-date business content.<\/p>\n Coordinate multiple specialized agents<\/strong> within unified workflows to handle complex, multi-step business processes. Agents can collaborate seamlessly to boost operational efficiency and accelerate decision-making across your organization.<\/p>\n Microsoft is consolidating Semantic Kernel and AutoGen into a unified agentic AI toolkit<\/strong>, making Semantic Kernel the primary framework for production-ready AI applications. As announced at Build 2025, this strategic merger brings AutoGen’s advanced multi-agent research capabilities into Semantic Kernel’s stable, enterprise-grade foundation\u2014delivering composable agent patterns with enhanced interoperability including Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocol support while maintaining the reliability needed for production deployments.<\/p>\n Access a centralized catalog<\/strong> of agent code samples with predefined instructions, actions, APIs, knowledge, and tools. Quickly bootstrap solutions for translation, sales, customer service, and domain-specific workflows.<\/p>\n Industry-focused healthcare agent tools<\/strong> support multi-disciplinary data workflows, helping clinicians streamline repetitive tasks and improve care delivery through intelligent automation.<\/p>\n The updated Azure AI Foundry VS Code extension<\/strong> now supports full agent CRUD operations, YAML IntelliSense, and direct deployment\u2014perform complete Azure AI Foundry workflows without leaving your IDE.<\/p>\n New trace agents<\/strong> functionality provides detailed visibility into agent threads, displaying the exact sequence of inputs and outputs for each agent run. Debug and monitor agent behavior with unprecedented transparency.<\/p>\n Major GitHub integration enhancements<\/strong> streamline AI-centric development workflows. Manage Azure AI Foundry functionalities directly within GitHub repositories for seamless code collaboration and agent deployment.<\/p>\n Automatically invoke agents<\/strong> in response to events like new emails or customer tickets using Azure Logic Apps integration. Enable real-time automation and responsiveness across your business processes.<\/p>\n New specialized tools include Bing Custom Search<\/strong> for grounding agents with web data and Morningstar integration<\/strong> for investment research\u2014expanding the capabilities of deployed agents across industries.<\/p>\n Azure AI Search introduces multi-vector field support<\/strong> for nested complex fields and scoring profiles integration with semantic ranking<\/strong>. Handle multimodal datasets, segment long documents into searchable chunks, and maintain precise relevance with enhanced ranking that carries scoring boosts through the final reranked results.<\/p>\n Reinforcement fine-tuning<\/strong> capabilities arrive for the o4-mini model, enabling highly tailored and efficient model customization for specialized reasoning tasks and agentic applications.<\/p>\n The Azure AI Foundry resource<\/strong> unifies agents, models, and tools under a single management grouping<\/strong> with built-in enterprise-readiness capabilities. Foundry projects<\/strong> within these resources are secure folders that enable developers to independently create new environments for exploring ideas and building prototypes while managing data in isolation. Features include integrated tracing & monitoring, agent and model-specific evaluation capabilities, and customizable enterprise setup configurations with new onboarding wizards<\/strong> that accelerate configuration with virtual network support and organizational policy compliance\u2014enabling developer self-service within established IT governance frameworks while making enterprise adoption faster and more secure.<\/p>\n Azure OpenAI now supports next-generation v1 APIs<\/strong> that eliminate the need to constantly update API versions. Developers can now use The Azure AI Foundry Projects client library<\/strong> provides a unified development experience that consolidates multiple client libraries into a single endpoint connection. Developers can now access Foundry Models (including Azure OpenAI), Agent Service, cloud evaluations, tracing capabilities, and external resource connections through one comprehensive SDK\u2014eliminating the complexity of managing separate libraries and enabling seamless integration across the entire AI development lifecycle. Recent updates to the Azure.AI.Agents.Persistent SDK (v1.1.0-beta.2) include Bing Custom Search tool<\/strong>, SharePoint grounding tool<\/strong>, Microsoft Fabric tool<\/strong>, and improved file upload handling for non-ASCII filenames.<\/p>\n Enhanced global training<\/strong> for fine-tuning enables distributed datasets and multi-region support for AI model customization\u2014lowering operational costs while improving performance even with smaller models.<\/p>\n The new Voice Live API<\/strong> enables developers to build low-latency, multimodal voice experiences<\/strong> with streamlined speech-to-speech conversations. Consolidate transcription, inference, and text-to-speech into a single API call via WebSocket connections, supporting real-time voice interactions with built-in noise suppression, echo cancellation, and function calling capabilities.<\/p>\n Safety leaderboards<\/strong> within Foundry model leaderboards help incorporate model safety as a first-class criterion<\/strong> alongside quality, cost, and throughput. Features include dedicated safety rankings using HarmBench benchmarks across 7 harmful content categories, quality-safety trade-off charts, and scenario-specific leaderboards for toxic content detection and sensitive domain knowledge\u2014essential for regulated industries like healthcare, education, and financial services.<\/p>\nlatest<\/code> and preview<\/code> endpoints, unified SDK consolidating multiple client libraries, and VS Code extension with full agent CRUD + YAML IntelliSense. APIs to know:<\/strong> Voice Live API (beta) for real-time speech interactions, enhanced Responses API with MCP tool calling, and new SDK updates including Bing Custom Search, SharePoint grounding, and Microsoft Fabric tools. Plus Model Safety Leaderboards help incorporate safety as a first-class criterion alongside performance. Building on Azure’s second consecutive year as a Gartner Leader<\/a> in data science and ML platforms, these updates pair technical excellence with proven organizational frameworks<\/a> that help development teams scale from prototype to production.<\/em><\/p>\n\ud83d\ude80 Join the Azure AI Foundry Developer Community<\/h2>\n
Models<\/h2>\n
o3-pro Enterprise Reasoning<\/h3>\n
Sora Video Generation<\/h3>\n
codex-mini Lightweight Coding Assistant<\/h3>\n
Grok 3 & Grok 3 Mini Now Available<\/h3>\n
DeepSeek-R1-0528 Advanced Reasoning<\/h3>\n
\nAgents<\/h2>\n
Model Context Protocol Support (Preview)<\/h3>\n
SharePoint Tool Integration (Preview)<\/h3>\n
Multi-Agent Workflows & Orchestration<\/h3>\n
Unified Semantic Kernel + AutoGen Runtime<\/h3>\n
Agent Catalog & Code Samples<\/h3>\n
Healthcare Agent Orchestrator<\/h3>\n
\nTools<\/h2>\n
Enhanced VS Code Extension<\/h3>\n
Trace Agents Debugging<\/h3>\n
Enhanced GitHub Development Platform<\/h3>\n
Logic Apps Agent Triggers<\/h3>\n
Agent-Specific Tools Expansion<\/h3>\n
Azure AI Search Multi-Vector & Semantic Enhancements<\/h3>\n
Reinforcement Fine-Tuning for o4-mini<\/h3>\n
\nPlatform (API, SDK, UI, and more)<\/h2>\n
New Azure AI Foundry Resource Type & Projects<\/h3>\n
Simplified API Version Management with
latest<\/code> and preview<\/code><\/h3>\nlatest<\/code> for stable features and preview<\/code> for cutting-edge capabilities\u2014no more monthly version updates or environment variable changes. The new approach also supports standard OpenAI client libraries with minimal code changes, making it easier to switch between OpenAI and Azure OpenAI implementations.<\/p>\nUnified Azure AI Foundry SDK<\/h3>\n
Global Fine-Tuning Training Capabilities<\/h3>\n
Voice Live API (Public Beta)<\/h3>\n
Model Safety Leaderboards (Public Preview)<\/h3>\n