LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks

Large language model (LLM) agents increasingly undertake long-horizon tasks that require sustained reasoning, tool use, and revision across many interdependent steps. However, existing agent harnesses maintain task execution, task state, and completion assessment within a growing context, making the state difficult to track and allowing incorrect self-assessments to propagate into later decisions. We reformulate long-horizon execution as a task-state management problem and propose LongHorizon- Harness, which maintains the task state explicitly outside execution and updates it only with facts independently verified from the environment. Its Manage-Execute-Audit (MEA) loop uses a manager to maintain the task state and determine the next subtask, a fresh-context executor to perform it, and a read-only auditor to verify the resulting environment state before the next round. A lightweight AgentAdapter supports interchangeable model and harness backends without modifying their native agent loops. LongHorizon-Harness improves Qwen 3.7-Plus from 51.8% to 80.7% on WeaveBench, from 69.7% to 77.2% on Terminal-Bench 2.1, and from 2.8% to 8.3% on OSWorld 2.0.

Introduction. Over the past few years, large language models (LLMs) have evolved from conversational models into the decision-making core of autonomous agents for software engineering (Yang et al., 2024; Wang et al., 2025; Ren et al., 2026; Ding et al., 2025), general-purpose assistance (Anthropic, 2025; OpenAI, 2025a;b; Anthropic, 2026), scientific discovery (Sun et al., 2025; Wei et al., 2025a), computer use (Anthropic, 2024; OpenAI, 2025c; Google DeepMind, 2025; Sager et al., 2026; Zhou et al., 2026; Zheng et al., 2026), and multimodal interaction (Zhang et al., 2023; Agashe et al., 2025). Across these applications, agents increasingly face long-horizon execution, which requires repeated reasoning, tool use, observation, and revision over many interdependent steps, sometimes across multiple context windows or sessions. The length of tasks that agents can complete increasingly determines how much work can be delegated to them.

Discussion / Conclusion. In this work, we introduce LongHorizon-Harness, a general framework for long-horizon agent execution that separates task-state management from environment interaction through a Manage–Execute–Audit loop. LongHorizon-Harness maintains progress as explicit, audited task state, executes each subtask in a fresh context, and carries only independently verified outcomes across rounds. Experiments on WeaveBench, OSWorld 2.0, and Terminal-Bench 2.1 show consistent improvements across hybrid GUI–CLI workflows, professional desktop tasks, pure command-line environments, and different model backbones. These results demonstrate that long-horizon agent capability is determined not only by the underlying model, but also by the harness that organizes, verifies, and converts its local capabilities into end-to-end task completion.