EvoCowork: A Benchmark for Real-World Office Tasks

arXiv:2610.14828 · cs.AI · Submitted 2026-10-10 · Read on arXiv

cs.AI

Submitted: 2026-10-10

Updated: 2026-10-10

Code: https://github.com/microsoft/STATE-Bench

Project page: https://microsoft.github.io/ppteval

Terminology

• Benchmark — 1× in this paper · explained in 2 episodes

Sources

• AdaRubric: Task-Adaptive Rubrics for Reliable LLM Agent Evaluation and Reward Learning
• ContinualSkillBench: Can LLM Agents Truly Evolve Their Capabilities?
• Designing Benchmarks for Knowledge Work
• Evo-Bench: Can Language Models Improve Agent Harness?
• Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses
• PPTArena: A Benchmark for PowerPoint Editing
• GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks
• Can LLM-as-a-Judge Reliably Verify Rubrics in Agentic Scenarios?
• Office Comprehension Benchmark
• S3Gym: Can LLMs Turn Self-Testing and Self-Judging into Self-Improvement?
• Workspace-Bench 1.0: Benchmarking AI Agents on Workspace Tasks with Large-Scale File Dependencies
• HarnessOpt-Bench: Evaluating LLMs at Harness Optimization
• APEX-Agents
• Rethinking the Evaluation of Harness Evolution for Agents
• OfficeBench: Benchmarking Language Agents across Multiple Applications for Office Automation
• OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments
• TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks
• PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents
• DarwinX: Evolving Agent Harnesses Through Natural Selection
• SEAGym: An Evaluation Environment for Self-Evolving LLM Agents

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems