Tool-use Failure
A failure to appropriately invoke or use a selected tool, set arguments, follow an execution order, interpret results, or recover from errors.
Example
A system selects a date-filtered search tool but calls it with the date field empty, then repeats the same invalid call after receiving an error.
Distinctions and cautions
Choosing an unsuitable tool is Tool-selection Failure, distinct from using a selected tool incorrectly. An outage of an external service is not automatically a model's Tool-use Failure. Persistently repeating an inappropriate invocation after a failed response may, however, qualify as a tool-use problem.
Evidence
View evidence →
API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs
- Title
- API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs
- Author / organization
- Li et al.
- Year
- 2023
- Source type / publication status
- EMNLP peer-reviewed research; published
- Supported concepts / claim context
- Planning, Tool-selection, Tool-use.
- Scope and limitations
- 73 tools, 314 dialogues and 753 calls; 2023 model set.
- Access / version note
- Published EMNLP.
- Recorded official URL / DOI / identifier
- DOI:
10.18653/v1/2023.emnlp-main.187(DOI resolver)
ACEBench
- Title
- ACEBench
- Author / organization
- Chen et al.
- Year
- 2025
- Source type / publication status
- Findings EMNLP peer-reviewed research; published
- Supported concepts / claim context
- Tool-selection and Tool-use errors.
- Scope and limitations
- Benchmark-defined tasks; taxonomy depends on benchmark design.
- Access / version note
- Published Findings EMNLP.
- Recorded official URL / DOI / identifier
- DOI:
10.18653/v1/2025.findings-emnlp.697(DOI resolver)
Robust Tool Use via Fission-GRPO
- Title
- Robust Tool Use via Fission-GRPO
- Author / organization
- Zhang et al.
- Year
- 2026
- Source type / publication status
- ACL peer-reviewed research; published
- Supported concepts / claim context
- Tool-use recovery; repeated invalid invocation.
- Scope and limitations
- Tested training methods and models; method-specific.
- Access / version note
- Published ACL.
- Recorded official URL / DOI / identifier
- DOI:
10.18653/v1/2026.acl-long.1880(DOI resolver)
EVOTOOL
- Title
- EVOTOOL
- Author / organization
- Yang et al.
- Year
- 2026
- Source type / publication status
- ACL peer-reviewed research; published
- Supported concepts / claim context
- Planner/selector/caller distinctions and cross-module propagation.
- Scope and limitations
- Modular architecture; not a universal architecture.
- Access / version note
- Published ACL.
- Recorded official URL / DOI / identifier
- DOI:
10.18653/v1/2026.acl-long.2016(DOI resolver)