Provider and attribution
Berkeley Function Calling Leaderboard (BFCL) V4
- Registry state
- registered
- Rights state
- approved
- Provider group
- bfcl-v4
Berkeley Function Calling Leaderboard (BFCL) V4 by the UC Berkeley Gorilla team (Apache License 2.0)
Release planning · no model scores
Berkeley Function Calling Leaderboard V4 evidence mapping
Exact V3 collection, facet, metric, lineage, identity and rights mapping for Berkeley Function Calling Leaderboard V4 from Berkeley Function Calling Leaderboard (BFCL) V4.
V3 evidence records
1 collection mapping
A repeated upstream benchmark can map to several collections. Each record below preserves its own task role, facets, capability atoms, subject identity, metrics, lineages and exclusions.
primary evidence · tool-use/bfcl-v4/berkeley-function-calling-leaderboard-v4
Tool Calling & Structured Output
Which models are best at tool calling and structured output?
- Upstream access
- machine-ready
- Eligibility
- runnable-now
- Registry
- registered
- Max task directness
- 90%
Collection facets
Function calling · objective · 25%
Select the right function and construct valid arguments.
Multi-turn tool use · hybrid · 25%
Maintain state and recover across dependent calls.
Structured schemas and values · objective · 25%
Produce schema-valid output with correct grounded values.
No-call and error recovery · hybrid · 25%
Decline irrelevant calls and recover safely from tool errors.
Capability atoms
function_callingmulti_turn_toolscall_decision
Subject identity
- Declared:
- harnessed-model
- Metric partitions:
- harnessed-model
- Separate partition required:
- no
- Identity projection:
- forbidden
Usable evidence lineages
bfcl-v4-f7cf735/harnessed-model
Exact metric contract
| Metric | Lineage | Subject | Protocol | Use |
|---|---|---|---|---|
bfcl_v4.single_turn_non_live_astBFCL V4 single-turn non-live AST accuracy | bfcl-v4-f7cf735 | harnessed-model | bfcl-v4@f7cf7359b7ac615a0b294831c5ba2bc95ee4a000 | usable |
bfcl_v4.single_turn_live_astBFCL V4 single-turn live AST accuracy | bfcl-v4-f7cf735 | harnessed-model | bfcl-v4@f7cf7359b7ac615a0b294831c5ba2bc95ee4a000 | usable |
bfcl_v4.multi_turnBFCL V4 multi-turn function-calling accuracy | bfcl-v4-f7cf735 | harnessed-model | bfcl-v4@f7cf7359b7ac615a0b294831c5ba2bc95ee4a000 | usable |
bfcl_v4.relevance_detectionBFCL V4 relevance detection accuracy | bfcl-v4-f7cf735 | harnessed-model | bfcl-v4@f7cf7359b7ac615a0b294831c5ba2bc95ee4a000 | usable |
bfcl_v4.irrelevance_detectionBFCL V4 irrelevance/no-call detection accuracy | bfcl-v4-f7cf735 | harnessed-model | bfcl-v4@f7cf7359b7ac615a0b294831c5ba2bc95ee4a000 | usable |
Exclusions and blockers
No mapping-level exclusion is recorded; the page-level release gate still applies.
Mapping note
Native and prompted tool calling are hard partitions.