OmegaUse-OfficeVal: Benchmarking LLM Agents on Long-Horizon Office-Suite Tasks with Economic Grounding

We introduce OmegaUse-OfficeVal, a benchmark for evaluating LLM agents on long-horizon Office-suite tasks with economic grounding, spanning word-processing documents, spreadsheets, presentations, and cross-file productivity tasks. The benchmark is constructed from authentic Office requests proposed by practitioners and adapted through a privacy-preserving process that removes sensitive information while preserving user intent, constraints, and natural request phrasing. OmegaUse-OfficeVal further provides task-level economic grounding through two complementary signals. Human labor time records the time required by human workers to complete the task without LLM assistance, while task price proxy estimates the market price of completing the task using explicit price signals when available and expert estimates otherwise. These annotations enable evaluation in terms of both task completion and the human effort and economic value associated with the work. We construct the benchmark through a multi-stage task adaptation pipeline and evaluate final deliverables using deterministic code-based verifiers. Experiments with frontier LLM agents and general-purpose computer-use or coding agents show that current systems make only partial progress on these realistic Office-suite workflows. OmegaUse-OfficeVal offers a reproducible, economically grounded testbed for measuring progress toward reliable LLM agents for everyday Office productivity.