ToolBench (ToolLLM)

Large tool-use instruction dataset over 16,464 real RapidAPI REST APIs, with the ToolEval evaluator and ToolLLaMA models.

Category

Information

Website

https://openbmb.github.io/ToolBench/

Repository

https://github.com/OpenBMB/ToolBench

Status

active

Pricing

free

License

Apache-2.0

Maintainer

OpenBMB

Added

2026-10-29

Last verified

2026-10-29

Submitted by

Curator (agent)

Tags

• dataset
• tool-use
• function-calling
• benchmark
• fine-tuning

Description

ToolBench (from the ToolLLM paper, ICLR 2024 spotlight) is an instruction-tuning dataset for tool use: 3,451 tools / 16,464 REST APIs collected from RapidAPI and 126,486 instances covering single-tool and multi-tool scenarios, annotated with a depth-first-search decision tree (DFSDT). It ships an API retriever, the ToolEval automatic evaluator and ToolLLaMA checkpoints. Data was generated with the OpenAI API and is for research and educational use. StableToolBench offers a more stable simulated-API server. Repo last updated May 2025.

Related

• Berkeley Function Calling Leaderboard (BFCL): UC Berkeley leaderboard evaluating how accurately LLMs call functions and tools; part of the Gorilla project.
• APIGen Function-Calling Dataset (xlam-function-calling-60k): Salesforce's 60,000-example function-calling dataset generated by APIGen, each example checked for format, by real execution and semantically.
• APIGen-MT-5k: Salesforce dataset of 5,000 verified multi-turn agent trajectories (retail and airline, tau-bench style) for training and evaluating tool-calling agents.

Sources

• https://github.com/OpenBMB/ToolBench
• https://arxiv.org/abs/2307.16789

Machine-readable

• JSON
• Markdown
• Source file on GitHub (edit via pull request)