---
title: "API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs"
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is arXiv non-exclusive distribution license (verified on arXiv abs page 2026-11-12); this pointer note is our own words and is CC BY 4.0; we link to the paper rather than reproduce it'
source: https://iinventedlight.com/library/notes/2304-08244-api-bank.md
source_license: "arXiv non-exclusive distribution license"
attribution: 'Pointer by iinventedlight (AI-assisted). Source: Minghao Li, Yingxiu Zhao, Bowen Yu, and 6 coauthors, API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs (arXiv non-exclusive distribution license). This note links to the paper rather than reproducing it.'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Pointer note only. Paper not reproduced. Abs page and license checked 2026-11-12.'
category: arxiv
topic_area: 'ai-research: tools'
tags: arxiv, tools, api, benchmarks, tool-use
arxiv_id: 2304.08244
alias: 2304-08244-api-bank
type: pointer
format: pointer
sources:
  - id: S1
    title: "API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs"
    publisher: arXiv
    authors: "Minghao Li and 8 coauthors (listed on the arXiv abs page)"
    url: https://arxiv.org/abs/2304.08244
    version_read: "v2 (25 Oct 2023)"
    accessed: 2026-11-12
    license: "arXiv non-exclusive distribution license"
---
# API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs

## Summary
[API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs](https://arxiv.org/abs/2304.08244) (2023) is a benchmark for language models that must plan, retrieve, and call APIs, not just talk about tools in the abstract.

What it introduced: a runnable evaluation system with 73 API tools and 314 annotated tool-use dialogues containing 753 API calls, plus a larger training set of 1,888 dialogues over 2,138 APIs in 1,000 domains. The authors train Lynx, a tool-augmented model starting from Alpaca, and they compare GPT-3, GPT-3.5, and GPT-4. They report that GPT-3.5 already uses tools better than GPT-3, that GPT-4 is stronger at planning, and that Lynx beats Alpaca by more than 26 points and approaches GPT-3.5. An error analysis lists remaining obstacles.

What it led to: a standard tool-use testbed used next to Toolformer, Gorilla, and later API-calling evals. Live notes: [2302-04761-toolformer](https://iinventedlight.com/library/notes/2302-04761-toolformer/), [2305-15334-gorilla](https://iinventedlight.com/library/notes/2305-15334-gorilla/), and [2307-16789-toollm](https://iinventedlight.com/library/notes/2307-16789-toollm/).

Where to read it: https://arxiv.org/abs/2304.08244 (v2, last revised 25 Oct 2023). EMNLP 2023.

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
