LiveClawBench: Benchmarking LLM Agents on Complex, Real-World Assistant Tasks

2026-03-20

arXiv.org

AI Summary

Existing benchmarks for large language model (LLM) agents struggle to simultaneously preserve the fidelity of real-world task distributions and execution semantics, often inadequately addressing practical challenges such as cross-service dependencies, state contamination, and implicit user intents. To bridge this gap, this work proposes LiveClawBench, a novel evaluation framework that enforces dual fidelity through a three-dimensional complexity model encompassing task structure, service interaction, and state evolution. The benchmark features 134 executable tasks across 10 domains and 22 simulated services, implemented within a full-stack, stateful, and reproducible execution environment. LiveClawBench provides an open-source dataset, a public leaderboard, and detailed trajectory logs, enabling controlled, fine-grained, and scalable assessment of LLM agents in realistic digital assistant scenarios.

Abstract

LLM-based agents are increasingly expected to handle real-world assistant tasks, yet existing benchmarks typically evaluate them under isolated sources of difficulty, such as a single environment or fully specified instructions. This leaves a substantial gap between current evaluation settings and the compositional challenges that arise in practical deployment. To address this gap, we introduce LiveClawBench, a benchmark to evaluate LLM agents on real-world assistant tasks. Based on an analysis of various real OpenClaw usage cases, we derive a Triple-Axis Complexity Framework that characterizes task difficulty along three dimensions: Environment Complexity, Cognitive Demand, and Runtime Adaptability. Guided by this framework, we construct a pilot benchmark with explicit complexity-factor annotations, covering real-world assistant tasks with compositional difficulty. Together, the framework and benchmark provide a principled foundation for evaluating LLM agents in realistic assistant settings, and establish a basis for future expansion across task domains and complexity axes. We are continuing to enrich our case collections to achieve more comprehensive domain and complexity coverage. The project page is at https://github.com/Mosi-AI/LiveClawBench.

Problem

• LLM agents
• benchmarking
• real-world tasks
• execution fidelity
• stateful environments

Innovation

• LiveClawBench
• LLM agents
• fidelity benchmarking
• stateful execution
• mock applications