benchmark:berkeley-function-calling

{
  "id": "benchmark:berkeley-function-calling",
  "_kind": "Benchmark",
  "_file": "benchmarks/benchmarks/benchmarks-tool-use.yaml",
  "_cluster": "benchmarks",
  "attributes": {
    "displayName": "Berkeley Function Calling Leaderboard (BFCL)",
    "homepageUrl": "https://gorilla.cs.berkeley.edu/leaderboard.html",
    "kind": "model-only",
    "targetsKind": "ModelVersion",
    "description": "BFCL (Berkeley Function Calling Leaderboard, from the Gorilla\nproject) is the canonical public leaderboard for LLM function-\ncalling and tool-use accuracy across simple, parallel, multiple,\nand live function-calling categories. Versions v1, v2 (live),\nand v3 (multi-turn / multi-step) have been released.\n"
  },
  "outgoingEdges": [
    {
      "from": "benchmark:berkeley-function-calling",
      "to": "skill-area:tool-use",
      "kind": "covers",
      "attributes": {
        "attributes": {
          "coverage": "full",
          "weight": 1
        }
      }
    }
  ],
  "incomingEdges": [
    {
      "from": "eval-run:bfcl.claude-sonnet-4-5.2025-09",
      "to": "benchmark:berkeley-function-calling",
      "kind": "for_benchmark"
    },
    {
      "from": "eval-run:bfcl.gpt-5.2025-08",
      "to": "benchmark:berkeley-function-calling",
      "kind": "for_benchmark"
    }
  ]
}