Berkeley Function Calling Leaderboard

What it measures

BFCL evaluates function-calling accuracy and broader agentic interactions. Its public leaderboard distinguishes native function calling from prompted formats and describes overall accuracy as an unweighted average of its subcategories.

This is a method guide. Undominated does not publish or blend this benchmark’s scores into its model ranking.

What the result does not establish

• An aggregate does not show which particular tool-use failure affects your workflow.
• Native function calling and a prompted workaround may need different validation and retry logic.
• Benchmark cost and latency describe the published evaluation; they are not quotes for your application.

Hold these conditions constant

• Benchmark revision and evaluated category
• Native function calling or prompt format
• Tool schemas, multi-turn state and error handling
• Harness revision, inference settings and observed failures

An evaluation for your team

Evaluate valid calls, missing arguments, irrelevant tools, failed tool responses and denied actions against your own schemas. Validate arguments before execution. Keep success, invalid-call rate, retries, cost and latency as separate observations.

Write down the task version, attempt count, accepted outcomes, observed spend and latency. Preserve failures in the denominator. A new model version or tool configuration deserves a new record.