Beyond Function Calling: Benchmarking Tool-Using Agents under Tool-Environment Unreliability

2026-06-24

AI Summary

This work addresses the limitation of existing tool-use evaluation frameworks, which overlook environmental unreliability and thus fail to assess agent robustness under realistic perturbations. To bridge this gap, we introduce ToolBench-X, a novel benchmark that systematically defines and implements five categories of recoverable tool-environment perturbations—such as specification drift and execution failures—and evaluates agents’ ability to diagnose issues and apply recovery strategies (e.g., retry, fallback, or cross-validation) during multi-step executable tasks across diverse domains. Our structured perturbation injection mechanism enables fine-grained assessment focused on task completion rather than mere function-call accuracy. Experimental results reveal a significant performance drop in current agents under perturbations, demonstrate that targeted prompting effectively enhances recovery capabilities, and show limited gains from test-time scaling. The benchmark includes an automated evaluation framework to support reproducible research.

Abstract

Large language models are increasingly deployed as agents that solve tasks by interacting with external tool environments. Although recent tool-use benchmarks increasingly cover complex task settings, they still largely assume clean, stable, and trustworthy tool environments, leaving tool-environment unreliability insufficiently examined. We introduce ToolBench-X, a benchmark for evaluating agents under recoverable reliability hazards. ToolBench-X contains executable multi-step tasks across diverse domains and sequential, parallel, and mixed workflows, each paired with deterministic tools and a canonical final answer for automatic evaluation. Starting from clean tool environments, ToolBench-X injects five structured hazard types: Specification Drift, Invocation Error, Execution Failure, Output Drift, and Cross-source Conflict. Crucially, each injected instance remains solvable through at least one valid recovery path, such as retrying, fallback, verification, or cross-checking. Experiments reveal a substantial reliability gap: agents that perform well with reliable tools often fail under recoverable hazards. Further analysis shows that failures are driven less by tool-use volume or inference budget than by limited hazard diagnosis and ineffective recovery. Targeted recovery hints recover many failed tasks, while test-time scaling yields more limited gains. These results suggest that tool-use evaluation should move beyond function-call accuracy toward task completion under unreliable tool environments. The code and data is available at https://github.com/Foreverskyou/ToolBench-X.

Problem

• tool-use
• unreliable environments
• agent reliability
• benchmarking
• recoverable hazards

Innovation

• tool-use reliability
• recoverable hazards
• agent robustness
• ToolBench-X
• tool-environment unreliability