reasoning benchmark
BFCL-v3 Leaderboard
BFCL-v3 evaluates large language models' function calling capabilities through multi-turn and multi-step interactions across 1000 test cases in domains including vehicle control, trading bots, travel booking, and file system management.
Updated Sep 5, 2026
Models19
Model coverage19
MetricScore
EvidenceB
BFCL-v3 Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelZA | Score77.80% | Percentile100.00% | Participants19 | EvidenceC | Evaluated |
| Rank02 | ModelZA | Score76.40% | Percentile94.44% | Participants19 | EvidenceC | Evaluated |
| Rank03 | ModelME | Score74.40% | Percentile88.89% | Participants19 | EvidenceC | Evaluated |
| Rank04 | ModelMI | Score72.00% | Percentile83.33% | Participants19 | EvidenceC | Evaluated |
| Rank05 | ModelAC | Score72.00% | Percentile77.78% | Participants19 | EvidenceC | Evaluated |
| Rank06 | ModelAC | Score71.90% | Percentile72.22% | Participants19 | EvidenceC | Evaluated |
| Rank07 | ModelAC | Score71.90% | Percentile66.67% | Participants19 | EvidenceC | Evaluated |
| Rank08 | ModelAC | Score71.70% | Percentile61.11% | Participants19 | EvidenceC | Evaluated |
| Rank09 | ModelAC | Score70.90% | Percentile55.56% | Participants19 | EvidenceC | Evaluated |
| Rank10 | ModelAC | Score70.30% | Percentile50.00% | Participants19 | EvidenceC | Evaluated |
| Rank11 | ModelAC | Score70.20% | Percentile44.44% | Participants19 | EvidenceC | Evaluated |
| Rank12 | ModelAC | Score68.70% | Percentile38.89% | Participants19 | EvidenceC | Evaluated |
| Rank13 | ModelAC | Score68.60% | Percentile33.33% | Participants19 | EvidenceC | Evaluated |
| Rank14 | ModelAC | Score67.70% | Percentile27.78% | Participants19 | EvidenceC | Evaluated |
| Rank15 | ModelAC | Score67.30% | Percentile22.22% | Participants19 | EvidenceC | Evaluated |
| Rank16 | ModelAC | Score66.30% | Percentile16.67% | Participants19 | EvidenceC | Evaluated |
| Rank17 | ModelAC | Score66.30% | Percentile11.11% | Participants19 | EvidenceC | Evaluated |
| Rank18 | ModelAC | Score63.30% | Percentile5.56% | Participants19 | EvidenceC | Evaluated |
| Rank19 | ModelAC | Score63.00% | Percentile0.00% | Participants19 | EvidenceC | Evaluated |
BFCL-v3 Highlights
The leading models and scores on this benchmark.
BFCL-v3 Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for BFCL-v3
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What is BFCL-v3?
What BFCL-v3 measures and how its scores work.
BFCL-v3, or Berkeley Function Calling Leaderboard v3, is a benchmark for evaluating large language models' function calling capabilities in multi-turn and multi-step interactions.
It measures contextual retention across turns, execution of multiple internal function calls, system state changes, and execution path correctness, reported as a Score ratio.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- BFCL-v3
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about BFCL-v3.
Which model scores highest on BFCL-v3?
GLM-4.5 is currently ranked first with 77.80%.
What are the top three models on BFCL-v3?
The current leaders are GLM-4.5 (77.80%), GLM-4.5-Air (76.40%), and LongCat-Flash-Thinking (74.40%).
Which BFCL-v3 model has the lowest official input price?
GLM-4.5-Air has the lowest matched official input price at $0.20 input / $1.1 output per 1M tokens.
Which models are fastest among BFCL-v3 results?
The fastest matched records are Qwen3 VL 4B Thinking (8.77 tok/s via DeepInfra).
Does the highest score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.