[MS] What's new in Azure AI Foundry | June 2025 - devamazonaws.blogspot.com
TL;DR
Key dev updates: Agent Service hits GA with Model Context Protocol (MCP) support—integrate 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: New unified Azure AI Foundry resource type consolidates management, VS Code extension gains full agent CRUD + YAML IntelliSense, and trace debugging gives you complete visibility into agent execution flows. APIs to know: Voice Live API (beta) for real-time speech interactions, enhanced Responses API with MCP tool calling, and reinforcement fine-tuning for o4-mini. Plus Foundry projects enable self-service dev environments within IT governance guardrails. Building on Azure's second consecutive year as a Gartner Leader in data science and ML platforms, these updates pair technical excellence with proven organizational frameworks that help development teams scale from prototype to production.🚀 Join the new Azure AI Foundry Developer Forum on GitHub
Connect with engineers and peers, ask questions, showcase your projects, vote in polls, and shape the roadmap—all in one place. Categories span Announcements, Product Feedback, Polls, Show-and-Tell, and Get Help. Bring your ideas, code, and curiosity!Models
o3-pro Enterprise Reasoning
o3-pro delivers the highest reasoning performance 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—designed for organizations tackling their most challenging reasoning problems with precision and depth.Sora Video Generation
Sora is now available in public preview 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—perfect for marketing campaigns, educational content, and creative workflows.codex-mini Lightweight Coding Assistant
The new codex-mini model joins Azure OpenAI as a lightweight, efficient option for code generation, completion, and programming assistance—ideal for embedded scenarios where performance and cost matter.Grok 3 & Grok 3 Mini Now GA
Previously in preview, Grok 3 and Grok 3 Mini are now generally available in Azure AI Foundry. These models excel at real-time conversational AI and general-purpose reasoning tasks with competitive performance and pricing.DeepSeek-R1-0528 Advanced Reasoning
DeepSeek-R1-0528 brings the latest evolution in reasoning-optimized models with improved depth of reasoning and inferencing capabilities. 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—ideal for intelligent agents, copilots, and research applications.Agents
Model Context Protocol Support (Preview)
Foundry Agent Service becomes first-class MCP clients with today's preview. Bring any remote MCP server—self-hosted or SaaS—and Azure AI Foundry automatically imports capabilities, keeps them updated, and routes calls through enterprise security. Think "USB-C for AI integrations"—the 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.SharePoint Tool Integration (Preview)
Seamlessly integrate agents with Microsoft SharePoint 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—maintaining enterprise-grade security while grounding AI responses in up-to-date business content.Pro Mode for Content Understanding
Accelerate decision-making with the new Pro mode in Azure AI Content Understanding, now available in Foundry. Announced at Microsoft Build, Pro mode enables fully agentic solutions by combining advanced content parsing with autonomous decision logic—ideal for scenarios requiring high-confidence, end-to-end automation.Multi-Agent Workflows & Orchestration
Coordinate multiple specialized agents 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.Unified Semantic Kernel + AutoGen Runtime
The merger of Semantic Kernel and AutoGen delivers a single, composable API for defining, chaining, and deploying agents locally or in the cloud with identical behavior—eliminating framework choice complexity.Agent Catalog & Code Samples
Access a centralized catalog of agent code samples with predefined instructions, actions, APIs, knowledge, and tools. Quickly bootstrap solutions for translation, sales, customer service, and domain-specific workflows.Healthcare Agent Orchestrator
Industry-focused healthcare agent tools support multi-disciplinary data workflows, helping clinicians streamline repetitive tasks and improve care delivery through intelligent automation.Tools
Enhanced VS Code Extension
The updated Azure AI Foundry VS Code extension now supports full agent CRUD operations, YAML IntelliSense, and direct deployment—perform complete Azure AI Foundry workflows without leaving your IDE.Trace Agents Debugging
New trace agents 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.Enhanced GitHub Development Platform
Major GitHub integration enhancements streamline AI-centric development workflows. Manage Azure AI Foundry functionalities directly within GitHub repositories for seamless code collaboration and agent deployment.Logic Apps Agent Triggers
Automatically invoke agents 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.Agent-Specific Tools Expansion
New specialized tools include Bing Custom Search for grounding agents with web data and Morningstar integration for investment research—expanding the capabilities of deployed agents across industries.Azure AI Search Multi-Vector & Semantic Enhancements
Azure AI Search introduces multi-vector field support for nested complex fields and scoring profiles integration with semantic ranking. 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.Reinforcement Fine-Tuning for o4-mini
Reinforcement fine-tuning capabilities arrive for the o4-mini model, enabling highly tailored and efficient model customization for specialized reasoning tasks and agentic applications.Platform
New Azure AI Foundry Resource Type
The Azure AI Foundry resource unifies agents, models, and tools under a single management grouping with built-in enterprise-readiness capabilities. Features include integrated tracing & monitoring, agent and model-specific evaluation capabilities, and customizable enterprise setup configurations tailored to your organizational policies—enabling consistent, efficient, and centrally governable AI development.Foundry Projects for Self-Service Development
Foundry projects are secure folders that enable developers to independently create new environments for exploring ideas and building prototypes while managing data in isolation. Projects share common settings like networking from their parent resource, enabling developer self-service within established IT governance frameworks.Unified Azure AI Foundry API
The new Azure AI Foundry API provides unified management and orchestration for agents, models, and development tools within a single resource grouping—streamlining the entire AI development lifecycle.Enhanced SDK with AgentOps
Updated SDK delivers built-in tracing, monitoring, and evaluation capabilities for agents and models. Benchmark performance, optimize solutions, and maintain compliance with enterprise-grade observability.Reimagined Portal Experience
The Azure AI Foundry portal introduces a unified dashboard for managing agents, models, and resources collectively. Enhanced self-serve features enable safe experimentation while maintaining central governance and compliance.Enterprise Setup Wizards
New enterprise onboarding wizards accelerate configuration with customizable setups, virtual network support, and organizational policy compliance—making enterprise adoption faster and more secure.Global Training Capabilities
Enhanced global training for fine-tuning enables distributed datasets and multi-region support for AI model customization—lowering operational costs while improving performance even with smaller models.Voice Live API (Public Beta)
The new Voice Live API enables developers to build low-latency, multimodal voice experiences 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.Comprehensive Documentation Updates
Extensive new documentation includes detailed changelogs, API/SDK guides, fine-tuning tutorials, and enterprise best practices—empowering both developers and administrators with complete learning resources.Model Safety Leaderboards (Public Preview)
Safety leaderboards within Foundry model leaderboards help incorporate model safety as a first-class criterion 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—essential for regulated industries like healthcare, education, and financial services.Enhanced Responses API with New Capabilities
The Responses API gains powerful new features including MCP tool calling support for interoperability with external agent frameworks, image generation integration with gpt-image-1, background mode for long-running tasks, enhanced reasoning controls with tunable effort levels, and improved text prompting with developer role hierarchy. These updates streamline agentic application development by combining simplicity with advanced capabilities in a single API.Happy building—and let us know what you ship with #AzureAIFoundry!
Post Updated on June 30, 2025 at 04:00PM
Thanks for reading
from devamazonaws.blogspot.com
Comments
Post a Comment