Berkeley Function Calling Leaderboard
Partial portExecutable evaluation of function selection, arguments, parallel calls, and multi-turn tool use.
3 case previewschatApache-2.0version bfcl v4
Run on your models
Requires the listed function tools, but does not require a machine sandbox.
What it measures
BFCL measures whether models can select functions and construct valid arguments across simple, parallel, and multi-turn scenarios. Tool schemas are part of each case rather than incidental benchmark configuration.
Trunchbull currently ports a curated set of cases and their immutable tool definitions. Full upstream parity still requires additional function mappings and the complete AST grading pipeline.
partial port
A curated Trunchbull port is available. Full BFCL v4 coverage and grader parity are still in progress.
Cases
3 in this release| id | case | runtime |
|---|---|---|
| bfcl/simple/weather-001 | Current weather lookupSelect a weather function and construct its location argument.tools: weather.get_current | chat |
| bfcl/parallel/flights-014 | Parallel flight statusIssue independent status calls for two flight numbers.tools: flights.get_status | chat |
| bfcl/multi-turn/calendar-008 | Resolve and scheduleCollect missing details before creating a calendar event.tools: calendar.find_slot, calendar.create_event | chat |
Tools
3 requested| tool | description | kind | required |
|---|---|---|---|
| weather.get_current | Returns current conditions for a named location. | function | yes |
| flights.get_status | Looks up the live status of a flight number. | function | yes |
| calendar.create_event | Creates an event from a resolved time range and attendees. | function | yes |