Microsoft Research

AI should work for everyone, wherever they are, and whatever language they speak. Introducing three interactive playbooks by Microsoft Research. These guides are for organizations building AI across languages, cultures, and markets. Meet Paza, Vibhasha, and Atlas. msft.it/6014QxJnm

AsgardBench evaluates whether embodied agents can revise their plans based on visual observations as tasks unfold. By focusing on perception-driven planning, it exposes key limitations and guides improvements in agent reliability. msft.it/6015QQ4fZ

GroundedPlanBench evaluates whether VLMs can plan actions and determine where they should occur. V2GP can improve both planning and spatial grounding, leading to more reliable robot behavior. Learn more: msft.it/6016QQQTg

Are machines truly intelligent? In Episode 1 of “The Shape of Things to Come,” technologists Subutai Ahmad & Nicolò Fusi join Microsoft’s Doug Burger to compare how large language models work with how the human brain learns & what it means for AI’s future. msft.it/6015QsiOZ

MicroLED datacenter networking, LLM-driven GPU design, real-time multimodal AI, agent memory and learning from experience, plus new benchmarks for interactive planning and a new podcast about the future of machine intelligence. msft.it/6016Qs9io

Think of datacenter networking like water moving through pipes. Traditional optical cables move data through a narrow, fast stream of light. While Microsoft’s new MicroLED system sends data through thousands of channels at once – more like a wide, slow-moving river carrying the same volume. The result: a new approach that researchers estimate could use about 50% less energy than mainstream laser-based optical cables, while supporting the massive data flows behind AI and cloud services. Learn how: msft.it/6001QoewP

With advances happening fast, what will the future look like? Microsoft Research Labs lead Doug Burger explores this question via the lens of machine intelligence, climate & more in the podcast “The Shape of Things to Come.” Full episodes coming soon. msft.it/6013QW7P5

Excited to showcase the science and people behind our GigaTIME paper, recently published in Cell. Paper: aka.ms/gigatime-paper Model: aka.ms/gigatime-model Code: aka.ms/gigatime-code Azure Foundry: labs.ai.azure.com/projects/gigat…

We’ve trained a multimodal AI model to turn routine pathology slides into spatial proteomics, with the potential to reduce time and cost while expanding access to cancer care.

As AI agents transition from simple chatbots to complex autonomous systems, finding and fixing their errors gets harder. AgentRx is an automated diagnostic framework that pinpoints critical failures and supports more transparent, resilient agentic systems: msft.it/6012QlyiM

With advances happening fast, what will the future look like? Microsoft research lead Doug Burger explores this question via the lens of machine intelligence, climate & more in the podcast series “The Shape of Things to Come.” Full episodes coming soon. msft.it/6018Qcv5i

Give this a second thought: You’re no longer just doing the work — you’re directing. In our latest episode, @sineadbovell explores how the future of work is taking shape, and what it means as AI systems start to feel less like tools and more like teammates. Watch all episodes of On Second Thought: msft.it/6015QcqMd

PlugMem transforms AI agents’ interaction histories into structured, reusable knowledge. It integrates with any agent, supports diverse tasks and memory types, and maximizes decision quality while significantly reducing memory token use: msft.it/6017Qc9vv

Multimodal reasoning with Phi-4-reasoning-vision, new work on scaling LLM inference, benchmarking AI agents in network operations, cinematic video generation, adaptive evaluation for LLMs, and using AI to improve individual and population health. msft.it/6013QiQgx

Vision-language models improve multimodal systems, but can make them slower, costlier, and harder to deploy. Learn how Phi-4-reasoning-vision-15B, a compact and fast multimodal reasoning model, blends strengths of different methods while reducing their limits: msft.it/6014Q5X0u

CORPGEN enables AI agents to manage dozens of interdependent tasks simultaneously in simulated workplace environments. It maintains performance under heavy multitasking, delivering up to 3.5x higher completion rates than leading baselines. msft.it/6015QbHoH

AI is no longer a future concept; it’s reshaping how we work right now. Explore how AI is rewriting the rules of work, leadership, and skills in this episode of On Second Thought. @sineadbovell sits down with Colette Stallbaumer, General Manager of M365 Copilot, to discuss what it means to move from doing the work to directing the work, why curiosity is becoming a critical competitive advantage, and how employees at every level (not just executives) can actively shape the future of work. From collaborating with AI agents to redefining roles and responsibilities, this episode breaks down what’s changing and how to prepare. Whether you’re a leader, knowledge worker, or simply curious about how AI will impact your career, this conversation offers practical insight into working with AI, not being replaced by it. Chapters: 02:45 What does the future org chart look light? 06:32 Are AI systems tools or something more? 11:24 How do we navigate the uncertainty about AI and work? Watch recent episodes of On Second Thought here: microsoft.com/en-us/research…

Long-term glass data storage advances, new work on transferable reasoning, multi-turn AI safety testing, multilingual AI design, and evaluating how models actually think. msft.it/6015QksvJ

Coming March 3 at 9:00 AM PT, our first Microsoft Research Forum episode of the year. New ARO, Dion2, Magentic Marketplace, OptiMind, Agent Lightning, and Healthbots. Register to watch: msft.it/6016Qn4hK