Vibeleaderboard
← Benchmark guide

Model results

Berkeley Function Calling Leaderboard

Choose a shared evaluation protocol to compare model scores. Versions, sources and configurations stay separate.

Atlas snapshot captured 10/19/2026, 10:20:28 AM UTC. This is an import timestamp, not the evaluation date. Coverage may be incomplete.

Reviewed recommendation and limits

Start with the relevant BFCL task category for function selection, arguments and multi-turn API use. Inspect category results before the overall score.

BFCL V4 with pinned evaluator revision and interface. Preserve single-turn, multi-turn, web-search and memory categories. Keep V3 and V4 totals separate.. Imported groups below may use different protocols; this review does not certify each run. AST matching, state checks and required-call checks measure different failure modes. Native function calling and prompted interfaces differ. The current V4 methods specify weighted categories, while the leaderboard summary says unweighted; preserve the exact revision and do not recompute a headline from that summary.

Publisher results Multi-turn graders V4 methods and aggregation
← All evaluation groups

Berkeley Function Calling Leaderboard

Source
Berkeley Function Calling Leaderboard (reviewed publisher results)
Version
V4
Measurement date
Not reported
Metric
% · Higher is better
Evaluation scope
Overall accuracy; configuration (FC); BFCL V4 weighted overall: Agentic 40%, Multi-Turn 30%, Live 10%, Non-Live 10%, Hallucination Measurement 10%; publisher-declared evaluation revision f7cf7359b7ac615a0b294831c5ba2bc95ee4a000, bfcl-eval 2025.12.17; publisher model-specific adapters and default request policy, not equal compute or fixed reasoning budgets; reasoning-model provider-default effort unspecified in published requests; raw execution responses, weight revisions, full launch settings, and per-run dates unverified; result date not supplied

Exact score ties share a position. Positions do not establish statistical significance. Missing metadata is shown as unreported; matching imported metadata is not independent verification of the protocol.

  1. #1Claude Opus 4.5

    anthropic/claude-opus-4-5

    77.47%

    View modelEvidence independent

  2. #2Claude Sonnet 4.5

    anthropic/claude-sonnet-4-5

    73.24%

    View modelEvidence independent

  3. #3GPT-5.2

    openai/gpt-5.2-20251211

    55.87%

    View modelEvidence independent

  4. #4GPT-5 Mini

    openai/gpt-5-mini-2025-08-07

    55.46%

    View modelEvidence independent

  5. #5GPT-4.1

    openai/gpt-4.1-2025-04-14

    53.96%

    View modelEvidence independent

  6. #6o4 Mini

    openai/o4-mini-2025-04-16

    53.24%

    View modelEvidence independent

  7. #7GPT-5 Nano

    openai/gpt-5-nano-2025-08-07

    51.45%

    View modelEvidence independent

  8. #8GPT-4.1 Mini

    openai/gpt-4.1-mini-2025-04-14

    50.45%

    View modelEvidence independent

  9. #9o3

    openai/o3-2025-04-16

    48.56%

    View modelEvidence independent

  10. #10Command A

    cohere/command-a-03-2025

    46.49%

    View modelEvidence independent

  11. #11GPT-4.1 Nano

    openai/gpt-4.1-nano-2025-04-14

    33.05%

    View modelEvidence independent

  12. #12Command R7B (12-2024)

    cohere/command-r7b-12-2024

    32.07%

    View modelEvidence independent

  13. #13Llama 3.3 70B Instruct

    meta/llama-3.3-70b-instruct

    31.9%

    View modelEvidence independent

  14. #14Llama 4 Scout

    meta/llama-4-scout-17b-16e-instruct

    28.13%

    View modelEvidence independent

  15. #15Llama 3.2 3B Instruct

    meta/llama-3.2-3b-instruct

    21.95%

    View modelEvidence independent

  16. #16Llama 3.2 1B Instruct

    meta/llama-3.2-1b-instruct

    10.82%

    View modelEvidence independent
Berkeley Function Calling Leaderboard model results | VibeLeaderboard