Berkeley Function Calling Leaderboard v4 (BFCL v4)
A function-calling benchmark for tool selection, schema adherence, and argument correctness.
Top models on BFCL v4 — November 13, 2026
As of November 13, 2026, BTL-3 leads the BFCL v4 leaderboard with 88.5% , followed by Atria Dawn Preview (77.0%) and Qwen3.7 Max (75.0%).
BTL-3
Bad Theory Labs
Reported by Bad Theory LabsAtria Dawn Preview
Shanghai Artificial Intelligence Laboratory
Reported by the labQwen3.7 Max
Alibaba
Reported by Alibaba26 modelsAgentic3% of Agentic reference weightCurrentUpdated November 13, 2026
Leaderboard (26 models)
None of the 26 results on this page has a documented run by the benchmark or an independent evaluator: 25 reported by the labs that built the models and 1 reported in another publisher’s document.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | BTL-3Bad Theory LabsReported by Bad Theory Labs | 88.5% | Reported by Bad Theory Labs | Not reported | Open |
| 2 | Atria Dawn PreviewShanghai Artificial Intelligence LaboratoryReported by the lab | 77.0% | Reported by the lab | Not reported | Open |
| 3 | Qwen3.7 MaxAlibabaReported by Alibaba | 75.0% | Reported by Alibaba | Not reported | Closed |
| 4 | BTL-4Bad Theory LabsReported by Bad Theory Labs | 73.5% | Reported by Bad Theory Labs | Not reported | Open |
| 5 | Ling 3.0 FlashInclusionAIReported by InclusionAI | 73.0% | Reported by InclusionAI | Not reported | Open |
| 6 | Qwen3.7 PlusAlibabaReported by Alibaba | 72.9% | Reported by Alibaba | Not reported | Closed |
| 7 | Spark-X2.6-4BSparkLLMReported by SparkLLM | 71.8% | Reported by SparkLLM | Not reported | Open |
| 8 | LFM3-1.7BLiquidAIReported by LiquidAI | 71.4% | Reported by LiquidAI | Not reported | Open |
| 9 | Pokee-Isaac 28BPokee AIReported by Pokee AI | 70.9% | Reported by Pokee AI | Not reported | Closed |
| 10 | Kimi K2.5Moonshot AIAnother lab’s comparison table | 68.3% | Another lab’s comparison table | Not reported | Open |
| 11 | MiniCPM5-2BOpenBMBReported by OpenBMB | 66.6% | Reported by OpenBMB | Not reported | Open |
| 12 | LFM2.5-24B-A2BLiquidAIReported by LiquidAI | 65.0% | Reported by LiquidAI | Not reported | Open |
| 13 | Granite 4.2 30BIBMReported by IBM | 61.4% | Reported by IBM | Not reported | Open |
| 14 | LLaDA2.2-flashInclusionAIReported by InclusionAI | 60.8% | Reported by InclusionAI | Not reported | Open |
| 15 | LFM2.5-2.6BLiquidAIReported by LiquidAI | 56.9% | Reported by LiquidAI | Not reported | Open |
| 16 | Granite 4.2 3BIBMReported by IBM | 52.4% | Reported by IBM | Not reported | Open |
| 17 | Granite 4.2 8BIBMReported by IBM | 52.4% | Reported by IBM | Not reported | Open |
| 18 | LFM2.5-8B-A1BLiquidAIReported by LiquidAI | 49.7% | Reported by LiquidAI | Not reported | Open |
| 19 | LLaDA2.2-miniInclusionAIReported by InclusionAI | 47.7% | Reported by InclusionAI | Not reported | Open |
| 20 | Mellum2-12B-A2.5B-ThinkingJetBrainsReported by JetBrains | 45.6% | Reported by JetBrains | Not reported | Open |
| 21 | Mellum2-12B-A2.5B-InstructJetBrainsReported by JetBrains | 44.2% | Reported by JetBrains | Not reported | Open |
| 22 | ZAYA1-8BZyphraReported by Zyphra | 39.2% | Reported by Zyphra | Not reported | Open |
| 23 | LFM2.5-VL-3BLiquidAIReported by LiquidAI | 32.5% | Reported by LiquidAI | Not reported | Open |
| 24 | MiniCPM5-1BOpenBMBReported by OpenBMB | 25.1% | Reported by OpenBMB | Not reported | Open |
| 25 | LFM2.5-VL-450MLiquidAIReported by LiquidAI | 21.1% | Reported by LiquidAI | Not reported | Open |
| 26 | LFM2.5-230MLiquidAIReported by LiquidAI | 21.0% | Reported by LiquidAI | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Results reported by model developersUnranked
Labs ran these themselves, with a metric, split or setting the table above does not use. They take no rank here.
| Model | Reported value | What was measured | Against the table above | Who ran it |
|---|---|---|---|---|
| Mellum2.1-12B-A2.5B-ThinkingJetBrains | 62.3%Macro-average of five subtasks (v1, v2, v3, web search, memory) | Thinking mode, greedy decodingJetBrains ran every model in the table with the same pipeline; not the official BFCL v4 overall weighting | Not comparable to the scores above | Reported by JetBrains |
A lab’s own result is shown for reference. It never takes a rank on this board.
How we label sourcesAccording to BenchLM.ai, BTL-3 leads the BFCL v4 benchmark with a score of 88.5%, followed by Atria Dawn Preview (77.0%) and Qwen3.7 Max (75.0%). There is significant spread across the leaderboard, making this benchmark effective at differentiating model capabilities.
26 models have been evaluated on BFCL v4. The benchmark falls in the Agentic category. BenchAlign v5.8 gives BFCL v4 3% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About BFCL v4
Year
2025
Tasks
Function-calling tasks
Format
Tool invocation and schema evaluation
Difficulty
Advanced tool use
BenchLM stores BFCL v4 as a function-calling benchmark.
Freshness and provenance
Version
BFCL v4 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does BFCL v4 measure?
A function-calling benchmark for tool selection, schema adherence, and argument correctness.
Which model scores highest on BFCL v4?
BTL-3 by Bad Theory Labs currently leads with a score of 88.5% on BFCL v4.
How many models are evaluated on BFCL v4?
26 AI models have published results on BFCL v4 in the BenchLM catalog.
Compare top models on BFCL v4
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.