Microsoft · 43 records

dateidtitleauthorsvenue↑
2026-11-042611.03634WebHarbor: Docking Real Websites for Evolving GUI Agent EnvironmentsZhaoyang Wang et al.arXiv24
2026-11-042611.02079ParetoSaddler: Pareto-Guided Optimization of Agent Harnesses for Quality and CostXingzhi Qian et al.arXiv18
2026-11-042610.40747Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved HarnessesXiaoyi Zhang et al.arXiv17
2026-10-272610.34886ThinkBit: Search Codewords, Recalibrate Scales, Select by ReasoningRhea Sanjay Sukthanker et al.arXiv2
2026-10-212610.26542Breaking Diffusion: Parallel Reasoning with Draft RefinementJaesang Yu et al.arXiv22
2026-10-192610.23194OfficeTown: Generative Organizations for Evolving Agent IntelligenceHongchen Wei et al.arXiv23
2026-10-152610.18072LensErase: A Foundation Model for Erase AnythingYunuo Chen et al.arXiv46
2026-10-142610.18860AesCode: Aesthetic Code Generation with Decoupled Cross-Modal RewardsXin Gao et al.arXiv43
2026-10-142610.18472Simplifying and Improving On-Policy Distillation for Few-Step DiffusionJiawei Zhang et al.arXiv27
2026-10-142610.14163AutoEnvScaling: Automating the Data Flywheel with Terminal AgentsSimon Yu et al.arXiv11
2026-10-092610.12230From Prompting to Composing: A Spatial Canvas Interface for Poster GenerationYitong Wang et al.arXiv13
2026-10-072610.05912MiniCorp: The Last Mile of the AI Agent FirmJingying Zeng et al.arXiv36
2026-10-062610.06361Capability-Driven Self-Evolution of Agent MemoryYaoqi Chen et al.arXiv15
2026-10-062610.02076LLM-as-Jev: LLMs Are Already Jev-Style Decision Models -- When and How to Fine-Tune ThemYinheng Li et al.arXiv14
2026-10-022610.00906ActiveSaddler: Automated Curriculum Learning for Agent Harness OptimizationSungho Park et al.arXiv59
2026-10-012609.35646Rubric Rewards from Item Response TheoryMilad Yazdani et al.arXiv11
2026-09-302609.37226Follow the Entities: A Corpus Map for Agentic SearchSoyeong Jeong et al.arXiv68
2026-09-232609.25804The Tasteful Agent: Measuring and Improving Taste in Long-Horizon TasksWenbo Pan et al.arXiv53
2026-09-182609.20511When EOS Tokens Disagree: Understanding Length Inflation in On-Policy DistillationYuxiao Yang et al.arXiv59
2026-09-182609.19671When2Think: Learning When and How Much to ReasonJaejun Shim et al.arXiv31
2026-09-032609.02812VibeVoice-ASR-Streaming Technical ReportYujie Tu et al.arXiv16
2026-08-312608.28444Sliding-window beats linear attentionAlexia Jolicoeur-Martineau et al.arXiv18
2026-08-262608.23041AutoSaddler: Automatic Harness Optimization with Durable Updates from Agent Execution TracesSungho Park et al.arXiv66
2026-08-252608.19741One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business WorkflowsZhuochun Li et al.arXiv8
2026-08-192608.17528Agent Lightning v1.0: Towards Harnessed Agentic RLZhiyuan He et al.arXiv35
2026-08-112608.08097OasisKV: Scaling In-Decode KV Cache Beyond HBM with Lookahead Sparse PrefetchingCan Xiao et al.arXiv23
2026-08-042608.00155AgentStream: How Well Do Self-Evolving LLM Agents Perform Under Streaming Tasks?Dong Yan et al.arXiv19
2026-07-292607.24904Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation ModelSenqiao Yang et al.arXiv38
2026-07-242607.21557OpenForgeRL: Train Harness-native Agents in Any EnvironmentXiao Yu et al.arXiv11
2026-07-222607.19064Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and EditingXinjie Zhang et al.arXiv76
2026-07-072607.04439ResearchStudio-Idea: An Evidence-Grounded Research-Ideation Skill Suite from ML Conference OutcomesQihao Zhao et al.arXiv59
2026-07-072607.04438ResearchStudio-Reel: Automate the Last Mile of Research from Paper to Poster, Video, and BlogLingao Xiao et al.arXiv56
2026-07-072606.31145SeKV: Resolution-Adaptive KV Cache with Hierarchical Semantic Memory for Long-Context LLM InferenceAmirhossein Abaskohi et al.arXiv6
2026-07-022606.31179HealthAgentBench: A Unified Benchmark Suite of Realistic Agentic Healthcare Environments for Challenging Frontier AI AgentsQianchu Liu et al.arXiv6
2026-07-022606.28430Building to the Test: Coding Agents Deliver What You Check, Not What You RequestedYanuo Ma et al.arXiv5
2026-06-182606.13802A Benchmark and Framework for Evaluating Next Action Predictions in SpreadsheetsTejas Agrawal et al.arXiv1
2026-06-162606.14066FastContext: Training Efficient Repository Explorer for Coding AgentsShaoqiu Zhang et al.arXiv77
2026-06-122606.09426WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid InterfacesWanli Li et al.arXiv46
2026-06-092606.05597AsyncWebRL: Efficient Multi-Step RL for Visual Web AgentsHao Bai et al.arXiv5
2026-06-022606.02031OpenWebRL: Demystifying Online Multi-turn Reinforcement Learning for Visual Web AgentsRui Yang et al.arXiv20
2026-05-252605.21573Lens: Rethinking Training Efficiency for Foundational Text-to-Image ModelsDong Chen et al.arXiv104
2026-05-132605.12501Covering Human Action Space for Computer Use: Data Synthesis and BenchmarkMiaosen Zhang et al.arXiv15
2026-05-012604.28181Synthetic Computers at Scale for Long-Horizon Productivity SimulationTao Ge et al.arXiv20