Microsoft Research has released Orchard, an open-source framework designed to enable scalable and cost-effective research into autonomous agents. The system centers on Orchard Env, a Kubernetes-based environment service that allows researchers to train and evaluate agents across diverse domains like software engineering, web navigation, and personal assistance without rebuilding infrastructure.
Microsoft Research has released Echoverse, a collection of twelve deep synthetic environments designed to train computer-use agents. The initiative includes ten domain-specific worlds and two capability-focused worlds, all built to maintain coherent state and behavioral fidelity.
Microsoft Research has introduced EvoLib, a framework that enables large language models to learn from their own experience during inference without requiring ground-truth labels or external feedback. Rather than storing raw experiences as static memories, EvoLib transforms them into reusable skills and reflective insights that are continually refined, consolidated, and reweighted.
Microsoft has open-sourced a branch of its SymCrypt cryptographic library containing formal specifications and proofs for the Rust implementations of ML-KEM and SHA-3. This release demonstrates a methodology where production algorithms are written in safe Rust and verified against standards-derived Lean models using the Aeneas toolchain.
Microsoft, Shanghai Jiao Tong University, Tongji University, and Fudan University researchers developed SkillOpt, a text-space optimizer that trains natural-language skill documents while keeping the target model frozen. The system uses an optimizer model to propose bounded edits based on scored rollouts, exporting the result as a single `best_skill.md` file.
Microsoft has released Mage-VL, an efficient 4B-parameter multimodal foundation model for image and video understanding that uses a codec-native approach to streamline visual processing. The system separates video streams into anchor (I) frames and predicted (P) frames, retaining only patches where the codec allocates bits to reduce visual token consumption by over 75%.
Microsoft AI has released MAI-Cyber-1-Flash, a cybersecurity-specialized model that runs inside the MDASH agentic scanning harness to improve vulnerability detection. The model is a sparse Mixture-of-Experts transformer with 137B total and 5B active parameters, derived from the MAI-Code-1-Flash lineage.
Microsoft has released Mage-Flow, a compact 4B-scale generative stack designed for efficient text-to-image generation and instruction-based image editing. The system achieves state-of-the-art-competitive quality through the co-design of a lightweight tokenizer, Mage-VAE, and a Native-Resolution Multimodal Diffusion Transformer (NR-MMDiT).
Microsoft Research AI Frontiers has released Fara1.5-27B, a multimodal computer use agent designed for web browsers. The model observes browser interfaces through screenshots and executes tasks by emitting structured tool calls such as clicking, typing, scrolling, and searching.
Microsoft introduces GFlowRL, a streamlined reinforcement learning algorithm that scales Generative Flow Networks to large language models by removing the auxiliary partition network. The method replaces the learned prompt-conditional partition function with an in-batch Monte Carlo estimate and adds importance-sampling correction and asymmetric flow-gap clipping for stability.
The authors present REGRIND, a minimalist retargeting-guided reinforcement learning pipeline that learns dexterous manipulation policies from a single human demonstration. The method retargets human hand-object motion to robot references while preserving spatial and contact relationships, then trains a residual RL policy in simulation to track object-centric keypoints.
Microsoft has uploaded the BitNet 0.6B embedding model to Hugging Face.
This research proposes a culturally and linguistically diverse (CALD) friendly AI-based detector for health misinformation, specifically addressing the lack of effective tools for low-resource languages like Bangla. The authors evaluate Small Language Models (SLMs) to overcome the limitations of Large Language Models in this context.