Berkeley Function Calling Leaderboard v3

test-set:bfcl-v3

TestSetbenchmarks/test-sets/test-sets.yaml·Open in Graph →

overview json graph

Attributes

displayName

Berkeley Function Calling Leaderboard v3

benchmarkId

benchmark:berkeley-function-calling

caseCount

4951

releasedAt

2024-09-19

composition

BFCL v3 extends the leaderboard with multi-turn and multi-step function-calling categories alongside the v1 simple/parallel/ multiple categories and the v2 "live" user-contributed prompts. The aggregate v3 test bank totals ~4,951 cases across all categories.

homepageUrl

https://gorilla.cs.berkeley.edu/leaderboard.html

description

BFCL v3 is the canonical multi-turn extension of the Berkeley Function Calling Leaderboard, released in September 2024. It is the standard public reference for LLM function-calling accuracy.

Outgoing edges (0)

None.

Incoming edges (2)

uses_test_set2

eval-run:bfcl.claude-sonnet-4-5.2025-09·EvalRun
eval-run:bfcl.gpt-5.2025-08·EvalRun