Inference
EN
Shishir G. Patil et al.

Berkeley Function Calling Leaderboard v4

A function-calling benchmark for tool selection, schema adherence, and argument correctness.

ModelsModels with a result.26
Top resultBest result on this benchmark.88.5%BTL-3
Top-3 spreadPoints from first to third.13.5 pts
YearYear of release.2025

Result and price

9 of 26 results

Results

26 results
#ModelResultThe score from the source.
01BTL-388.5%
02Atria Dawn Preview77.0%
03Qwen3.7 Max75.0%
04BTL-473.5%
05Ling 3.0 Flash73.0%
06Qwen3.7 Plus72.9%
07Spark-X2.6-4B71.8%
08LFM3-1.7B71.4%
09Pokee-Isaac 28B70.9%
10Kimi K2.568.3%
11MiniCPM5-2B66.6%
12LFM2.5-24B-A2B65.0%
13Granite 4.2 30B61.4%
14LLaDA2.2-flash60.8%
15LFM2.5-2.6B56.9%
16Granite 4.2 3B52.4%
16Granite 4.2 8B52.4%
18LFM2.5-8B-A1B49.7%
19LLaDA2.2-mini47.7%
20Mellum2-12B-A2.5B-Thinking45.6%
21Mellum2-12B-A2.5B-Instruct44.2%
22ZAYA1-8B39.2%
23LFM2.5-VL-3B32.5%
24MiniCPM5-1B25.1%
25LFM2.5-VL-450M21.1%
26LFM2.5-230M21.0%