Evaluation toolEvaluation tool

Berkeley Function Calling Leaderboard (BFCL)

Version 2026.3.23

Maintained by Gorilla project, UC Berkeley6 4 5

The Berkeley Function Calling Leaderboard (BFCL) is a benchmark and live leaderboard from UC Berkeley's Gorilla project for evaluating how well language models call functions and tools. It checks serial and parallel function calls in several programming languages using abstract syntax tree (AST) evaluation. Later versions add multi-turn and multi-step scenarios and, in V4, agentic tasks such as web search, memory management, and format sensitivity.6 1 4 5

Last reviewedSep 29, 2026Entry updated Sep 29, 2026Documented release Mar 23, 2026

Availability and license

Overall availability

Public

Code on GitHub under the Apache License 2.0 and installable from PyPI as bfcl-eval; the dataset is on Hugging Face under Apache 2.0.1 2 3 5

Availability is separate from permission: read the license before using or redistributing.

Apache License 2.0 (external site: raw.githubusercontent.com)2
Applies to: Code · SPDX Apache-2.0 · on the rubric’s OSI-approved list
Apache License 2.0 (external site: huggingface.co)5
Applies to: Data · SPDX Apache-2.0 · on the rubric’s OSI-approved list

Public materials checklist

Items for a evaluation tool under USASI rubric v0.1. Unknown means unassessed or insufficient evidence.

Public materials checklist for Berkeley Function Calling Leaderboard (BFCL)
Item	Status	Notes and evidence
CodeIs the evaluation code published?	Public	Response-generation and evaluation code is in the berkeley-function-call-leaderboard directory of the Gorilla repository.1 2
Tasks / dataAre the tasks or test data available?	Public	Published on Hugging Face as JSON files, one per test category. The card says the files are not compatible with the Hugging Face datasets load_dataset method; the card was last updated for V3.5
MethodologyIs the method for scoring described?	Public	The ICML 2025 paper describes AST-based checking, the mix of expert-curated and user-contributed functions, and the stateful multi-step agentic evaluation. Release blog posts cover each version.6 1
ReproducibilityAre instructions for reproducing results published?	Public	The README documents installation, response generation, and evaluation. The leaderboard names the code commit used for evaluation and makes model responses available.1 4
LimitationsAre known limitations documented?	Unknown	Not assessed.

What it is useful for

Evaluating function calling in API-hosted models and in locally hosted open models served through vLLM or SGLang, and comparing models on the public leaderboard, which also reports cost and latency.1 4

Run and use notes

Documented facts only. No hardware or performance claims are made without a cited source and stated assumptions.

The README says to install the PyPI package as bfcl-eval, not the unrelated bfcl package. PyPI installations must set the BFCL_PROJECT_ROOT environment variable, and the web-search category needs a SerpAPI key.1

Organization context

University of California, Berkeley

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S.-governed project

The ICML 2025 paper describing BFCL lists the University of California, Berkeley as the affiliation of all its authors. The leaderboard is hosted at gorilla.cs.berkeley.edu, and the dataset is published by the Hugging Face organization "Gorilla LLM (UC Berkeley)". The code sits in the Gorilla repository, hosted under an individual GitHub account, and no separate legal entity for the project was found. Eligibility rests on this documented UC Berkeley maintenance. UC Berkeley is a campus of the University of California, which is governed by The Regents under the California Constitution.6 4 5 1 7

Assessed Sep 29, 2026

Sources

1.
Berkeley Function Calling Leaderboard README (external site: raw.githubusercontent.com)
Gorilla project, UC Berkeley · Repository · accessed Sep 29, 2026
2.
Gorilla LICENSE (external site: raw.githubusercontent.com)
Gorilla project, UC Berkeley · License · accessed Sep 29, 2026
3.
bfcl-eval on PyPI (external site: pypi.org)
Python Package Index · Documentation · published Mar 23, 2026 · accessed Sep 29, 2026
4.
Berkeley Function Calling Leaderboard (external site: gorilla.cs.berkeley.edu)
UC Berkeley Gorilla project · Official page · accessed Sep 29, 2026
5.
gorilla-llm/Berkeley-Function-Calling-Leaderboard dataset card (external site: huggingface.co)
Gorilla LLM (UC Berkeley) · Dataset card · accessed Sep 29, 2026
6.
The Berkeley Function Calling Leaderboard (BFCL): From Tool Use to Agentic Evaluation of Large Language Models (PDF) (external site: raw.githubusercontent.com)
Proceedings of Machine Learning Research (ICML 2025) · Paper · published 2025 · accessed Sep 29, 2026
7.
About the Regents | Board of Regents (external site: regents.universityofcalifornia.edu)
University of California Board of Regents · Official page · accessed Sep 29, 2026

Report a correction Edit this entry

This listing is not an endorsement, a safety assessment, or a federal approval.