What is BFCL?

BFCL, the Berkeley Function Calling Leaderboard, tests whether a model picks the right function, fills in the right arguments, and calls nothing when nothing fits. Built and maintained by UC Berkeley’s Gorilla project, it isn’t a single frozen run: the leaderboard says it’s “updated periodically,” and its latest update, April 12, 2026, is published against a specific GitHub commit so anyone can reproduce the exact results behind it.

That matters because a model’s rank on BFCL can move between updates as new checkpoints get tested, so a number quoted from months ago is not the same claim as the live leaderboard. BFCL’s test cases and scoring code are public on GitHub and Hugging Face, so a team can run the same checks against its own tool definitions instead of reading the public leaderboard as a verdict on their setup. Which benchmark to use for tool calling specifically covers how BFCL stacks up against the alternatives.

sources

keep reading

More on this.

Does a high BFCL score mean an agent will work in production?Not fully. BFCL tests curated requests against mock backends that never time out, rate-limit, or return malformed data the way real production tools do.answer →Does BFCL's own test data have grading errors?Yes. Epoch AI's review found a grading defect in 24 of 50 sampled BFCL v4 questions, 48%, more than double its own threshold for a 'Flawed' verdict.answer →What changed in BFCL's V4?V4 cut single-turn testing from two-thirds of the score to a fifth and gave the freed weight to a new Agentic category: web search, memory, and format sensitivity.answer →Everything on bfclThe definition, and every question answered under bfcl.concept →

Try Tessary on Cloud.

start for freeor self-host

free · 10,000 traces a month · 1 GB · 30-day retention · $10 one-time credit for triage and RCA · no credit card