[MS] What’s new in Azure AI Foundry | April 2025 - devamazonaws.blogspot.com

TL;DR

Long-context GPT-4.1, GPT-image-1, new o-series reasoning and GPT-4o audio models headline this month’s releases. On the agent side we get cross-cloud A2A, BYO thread storage, an MCP server starter, and a turnkey AI Red Team. Developers also gain a VS Code extension, richer evaluation metrics, persistent memory via Mem0, a full RAG demo suite, and new Content Understanding & Document Intelligence endpoints—everything you need to build, test, and ship safer GenAI apps on a single platform.

Join the new Azure AI Foundry Developer Forum on GitHub

We launched the new GitHub Discussions Developer Forum last week and we're inviting you to connect with engineers and peers to ask questions, showcase your projects, vote in polls, and shape the roadmap—all in one place. Bring your ideas, code, and curiosity! [cta-button text="Open Discussions" url="https://ift.tt/0rfNjWX" color="btn-primary"]

Models

GPT-4.1 One-Million-Token Context

GPT-4.1 (and its nano/mini variants) lifts Azure’s context ceiling to 1 million tokens, letting you pass entire codebases or multi-gigabyte corpora in one shot, while retaining GPT-4-class reasoning and function calling. That means fewer chunk-and-stitch hacks, simpler prompts, and major latency savings for large-document RAG or full-repo code reviews. Learn how to call it with the Responses API
[cta-button text="Learn more" url="https://ift.tt/UWKNFVi" color="btn-secondary"]

GPT-image-1 Text-&-Image Generation

GPT-image-1 arrives in limited preview with sharper fidelity, reliable text rendering, editing / in-painting, and image-as-input support—so you can build marketing creatives, design mocks, and visual KB answers directly in Foundry using the same REST patterns as DALLE 3. [caption id="attachment_315" align="aligncenter" width="400"]foundry happy building wp image Prompt: A bustling metallurgical foundry is portrayed in an isometric illustration. The scene features developers operating computers, typing into keyboards, speaking into microphones and overseeing an "ai agent factory" creation process. The environment is filled with datacenter elements such as long rows of server racks, neatly organized networking cables, and large cooling apparatuses.​
Above the main workspace, a prominent banner stretches across the foundry floor, displaying the words "Happy Building" in bold, industrial-style lettering. The sign adds a touch of positivity and motivation amidst the intense industrial setting.[/caption] Happy Building!
[cta-button text="Generate images" url="https://ift.tt/YeUETI4" color="btn-secondary"]

o4-mini & o3 Reasoning Models

Need faster, cheaper reasoning? The new o-series pairs GPT-4-level logical depth with lower latency and aggressive pricing, making them ideal for agent planning, re-ranking, or embedded analytics where every millisecond (and penny) counts. [caption id="attachment_316" align="aligncenter" width="1024"]A graph showing improvements reasoning models demonstrate across challenging academic benchmarks such as GPQA Diamond, Codeforces, and AIME 2024. Source: https://openai.com/index/introducing-o3-and-o4-mini/[/caption]
[cta-button text="Learn more" url="https://ift.tt/cLfkPJh" color="btn-secondary"]

GPT-4o Audio (Transcribe & TTS)

gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-mini-tts bring high-quality speech-to-text and controllable text-to-speech to Azure. Stream captions, build multilingual voice bots, or generate audio replies—all via familiar /audio and /realtime endpoints. [caption id="attachment_318" align="aligncenter" width="1280"]Demonstration of the Azure OpenAI TTS Soundboard Demonstration of the Azure OpenAI TTS Soundboard[/caption] [cta-button text="Get started" url="https://ift.tt/y3HW5fX" color="btn-secondary"]

Agents

Semantic Kernel + A2A Interop

A new plug-in teaches Semantic Kernel to speak Google’s Agent-to-Agent JSON-RPC protocol, enabling secure cross-cloud agent collaboration—exchange context, not credentials, and orchestrate multi-modal workflows spanning Azure, GCP, and OSS runtimes. [cta-button text="Learn more" url="https://ift.tt/crtu27v" color="btn-secondary"]

MCP Server Starter (Typescript)

Spin up an MCP-compliant server in minutes; the template wires Azure AI Agents to Claude Desktop (or any MCP client) via standard JSON messages—no bespoke glue code required. [cta-button text="Get started" url="https://ift.tt/LpQB1w9" color="btn-secondary"]

BYO Thread Storage + Monitor

Agent Service now lets you store conversation threads in your own Cosmos DB and surfaces run metrics in Azure Monitor—boosting data residency compliance and giving SREs first-class observability out of the box. [cta-button text="Quickstart" url="https://ift.tt/0JLCx6s" color="btn-secondary"]

AI Red Teaming Agent (Preview)

Built atop Microsoft’s PyRIT toolkit, this agent fires automated jailbreak and prompt-injection probes at your models, scores Attack Success Rate, and logs findings into Foundry dashboards—making shift-left safety a one-command reality. [video width="1620" height="1080" mp4="https://devblogs.microsoft.com/foundry/wp-content/uploads/sites/89/2025/04/AI-red-teaming-agent_final_blog_asset-1.mp4"][/video] [cta-button text="Learn more" url="https://ift.tt/gChRDzZ" color="btn-secondary"]

Tools

VS Code Foundry Extension

Test models, deploy agents, and copy sample code without leaving VS Code—goodbye portal context-switching, hello faster inner loops. [video width="1920" height="1080" mp4="https://devblogs.microsoft.com/foundry/wp-content/uploads/sites/89/2025/04/25_66_Agent_v2.mp4"][/video] [cta-button text="Learn more" url="https://ift.tt/tBapCWm" color="btn-secondary"]

Quality & Safety Evaluators

Four new quality metrics (intent-resolution, tool-call accuracy, task adherence, completeness) plus code-vulnerability and ungrounded-attribute safety checks plug straight into CI/CD so every build ships with score gates. [video width="2160" height="1440" mp4="https://devblogs.microsoft.com/foundry/wp-content/uploads/sites/89/2026/04/eval-metrics-for-agents-2.mp4"][/video] [cta-button text="Learn more" url="https://ift.tt/q192pTZ" color="btn-secondary"]

Mem0 Persistent Memory Layer

Mem0 + Azure AI Search lets assistants remember user details across sessions via semantic retrieval—boosting personalization without extra infra. [cta-button text="Get started" url="https://ift.tt/EVc2wZG" color="btn-secondary"]

Content Understanding 2024-12-01 Preview

The new API adds generative & classification fields, faster video segmentation, and multi-analyzer docs, producing structured JSON ready for LLM ingestion across docs, audio, and video. [cta-button text="Read the docs" url="https://ift.tt/OQVLSU6" color="btn-secondary"]

Document Intelligence v4.0 Container

Run the Layout model on-prem or at the edge via new v4.0 containers—perfect for air-gapped PDF/OCR scenarios that need local processing yet Azure-compatible APIs. [cta-button text="Read the docs" url="https://ift.tt/Zih5Fqo" color="btn-secondary"]
Happy building—and let us know what you ship with #AzureAIFoundry!
Post Updated on April 29, 2025 at 04:30PM
Thanks for reading
from devamazonaws.blogspot.com

Comments

Post a Comment

Popular posts from this blog

Scenarios capability now generally available for Amazon Q in QuickSight - devamazonaws.blogspot.com

Today, AWS announces the general availability of the scenarios capability of Amazon Q in QuickSight. Amazon Q guides you through data analysis by uncovering hidden trends, making recommendations for your business, and intelligently suggesting next steps for deeper exploration—all in response to natural language interactions. Now anyone can explore past trends, forecast future scenarios, and model solutions without needing specialized skill, analyst support, or manual manipulation of data in spreadsheets. With its intuitive interface and step-by-step guidance, the scenarios capability of Amazon Q in QuickSight helps users perform complex data analysis up to 10x faster than spreadsheets. Whether you're optimizing marketing budgets, streamlining supply chains, or analyzing investments, Amazon Q makes advanced data analysis accessible so you can make data-driven decisions across your organization. This capability is accessible from any Amazon QuickSight dashboard, so you can move seam...
Read more

[MS] Introducing Pull Request Annotation for CodeQL and Dependency Scanning in GitHub Advanced Security for Azure DevOps - devamazonaws.blogspot.com

Image
In the world of software development, security is paramount. As developers, we strive to write clean, efficient, and most importantly, secure code. GitHub Advanced Security for Azure DevOps has always been at the forefront of providing tools that make it easier to build and release high-quality software. Today, we’re excited to announce a new feature release that will take your code security to the next level: PR (Pull Request) Annotation for CodeQL and Dependency Scanning. PR Annotation - What Does it Mean for You? Pull Request Annotation brings security insights directly into your development workflow. Here’s how it works: When you raise a Pull Request, CodeQL runs automatically, scanning for any potential security issues. Dependency Scanning concurrently checks for vulnerabilities in packages or libraries you might be using. If any issues are found, they are reported directly on the PR interface as annotations. Developers can see exactly what the problem is and where it occu...
Read more

AWS Console Mobile Application adds support for Amazon Lightsail - devamazonaws.blogspot.com

AWS customers can now access Amazon Lightsail from within the AWS Console Mobile App to monitor and manage Lightsail instances, containers, databases, network, storage, snapshots, domains and DNS while on the go. Visit the Services tab in the AWS Console Mobile App and select Lightsail to get started. The AWS Console Mobile App enables AWS customers monitor and manage a select set of resources and receive push notifications to stay informed and connected with their AWS resources while on-the-go. The sign-in process supports biometrics authentication, making access to AWS resources simple, secure, and quick. Lightsail offers easy-to-use virtual private server (VPS) instances, storage, databases, and more for a cost-effective monthly price. For AWS services not available natively, customers can access the AWS Management Console via an in-app browser to access service pages without additional authentication, manual navigation, or need to switch from the app to a browser. Visit the AWS...
Read more