Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

Berkeley Function Calling Leaderboard (BFCL) V4

Choosing the right function and arguments in one turn, on BFCL's curated set.

Data older than 90 days: last updated 16 Dec 2025

Results dated
16 Dec 2025
Results
109 configurations of 81 models
Unit
% correct
Licence
Apache License 2.0

Berkeley Function Calling Leaderboard (BFCL) V4 in brief

What it measures. Berkeley Function Calling Leaderboard (BFCL) V4 is a third-party benchmark, shown here with credit. Measured here: Choosing the right function and arguments in one turn, on BFCL's curated set.

Latest results. , 81 models ranked.

Top 3 on Single-turn calls (curated) (% correct, higher is better, each model at its best setting):

  1. Gemini 3 Pro Preview (Google): 90.7%
  2. Qwen3-235B-A22B-Instruct-2507 (Alibaba): 90.3%
  3. Qwen3-32B (Alibaba): 90.3%

How to cite. Berkeley Function Calling Leaderboard (BFCL) V4, via Spring Prompt, results dated 16 December 2025. https://springprompt.com/benchmarks/bfcl These results belong to Berkeley Function Calling Leaderboard (BFCL) V4, not to us. Licence: Apache License 2.0. Credit Berkeley Function Calling Leaderboard (BFCL) V4 first.

Single-turn calls (curated): GPT-5.2

Top 15 of 81 results · % correct, higher is better. Choose a model to highlight it.Clear highlight

  1. 1 Gemini 3 Pro PreviewPromptingGoogle 90.7%
  2. 2 Qwen3-235B-A22B-Instruct-2507PromptingAlibaba 90.3%
  3. 2 Qwen3-32BPromptingAlibaba 90.3%
  4. 4 Mistral Small 3.2 24BPromptingMistral AI 89.7%
  5. 4 Claude Opus 4.5PromptingAnthropic 89.7%
  6. 6 xLAM-2-32b-fc-rFunction callingSalesforce 89.6%
  7. 7 Qwen3-14BPromptingAlibaba 89.5%
  8. 8 Llama 4 ScoutFunction callingMeta 89.4%
  9. 9 Qwen3-30B-A3B-Instruct-2507PromptingAlibaba 88.9%
  10. 9 Arch-Agent-32BMode not statedKatanemo 88.9%
  11. 11 GPT-4.1PromptingOpenAI 88.7%
  12. 11 Claude Sonnet 4.5Function callingAnthropic 88.7%
  13. 11 Llama 4 Maverick 17B 128E Instruct (FP8)Function callingMeta 88.7%
  14. 14 Qwen3-8BPromptingAlibaba 88.6%
  15. 15 Mistral NemoPromptingMistral AI 88.5%
  16. 52 GPT-5.2Function callingOpenAI 81.9%

Full results

BFCL: single-turn calls (curated), % correct, higher is better
#ModelSingle-turn calls (curated)
% correct, higher is better
Price
$ per million tokens, in / out
1 Gemini 3 Pro PreviewPromptingGoogle · best of 2 settings
90.7%
–
2 Qwen3-235B-A22B-Instruct-2507PromptingAlibaba · best of 2 settings
90.3%
$0.15 / $0.75
2 Qwen3-32BPromptingAlibaba · best of 2 settings
90.3%
$0.14 / $0.40
4 Mistral Small 3.2 24BPromptingMistral AI · best of 2 settings
89.7%
$0.094 / $0.25
4 Claude Opus 4.5PromptingAnthropic · best of 2 settings
89.7%
$5 / $25
6 xLAM-2-32b-fc-rFunction callingSalesforce
89.6%
–
7 Qwen3-14BPromptingAlibaba · best of 2 settings
89.5%
$0.10 / $0.22
8 Llama 4 ScoutFunction callingMeta
89.4%
$0.10 / $0.30
9 Qwen3-30B-A3B-Instruct-2507PromptingAlibaba · best of 2 settings
88.9%
$0.09 / $0.30
9 Arch-Agent-32BMode not statedKatanemo
88.9%
–
11 GPT-4.1PromptingOpenAI · best of 2 settings
88.7%
$2 / $8
11 Claude Sonnet 4.5Function callingAnthropic · best of 2 settings
88.7%
$3 / $15
11 Llama 4 Maverick 17B 128E Instruct (FP8)Function callingMeta
88.7%
–
14 Qwen3-8BPromptingAlibaba · best of 2 settings
88.6%
–
15 Mistral NemoPromptingMistral AI · best of 2 settings
88.5%
$0.029 / $0.03
16 xLAM-2-70b-fc-rFunction callingSalesforce
88.4%
–
17 Grok 4.1 Fast (reasoning)Function callingxAI
88.3%
–
18 Grok 4.1 Fast (non-reasoning)Function callingxAI
88.1%
–
18 Gemini 2.5 FlashPromptingGoogle · best of 2 settings
88.1%
$0.30 / $2.50
20 Llama 3.3 70B InstructFunction callingMeta
88.0%
$0.71 / $0.71
21 Qwen3-4B-Instruct-2507Function callingAlibaba · best of 2 settings
87.9%
–
22 Command AFunction callingCohere
87.6%
$2.50 / $10
22 GLM-4.6 (reasoning on)Function callingZ.ai
87.6%
$0.50 / $2
24 Palmyra X 004Function callingWriter
87.5%
–
25 Gemma 3 27BPromptingGoogle
87.2%
$0.08 / $0.35
26 ToolACE-2-8BFunction callingHuawei Noah's Ark Lab and USTC
87.1%
–
27 Nova 2 LiteFunction callingAmazon
87.0%
$0.30 / $2.50
28 Arch-Agent-3BMode not statedKatanemo
86.7%
–
29 Gemini 2.5 Flash-LiteFunction callingGoogle · best of 2 settings
86.6%
$0.10 / $0.40
29 Nova Pro 1.0Function callingAmazon
86.6%
$0.80 / $3.20
31 Claude Haiku 4.5Function callingAnthropic · best of 2 settings
86.5%
$1 / $5
32 Command A ReasoningFunction callingCohere
86.3%
–
33 DeepSeek-V3.2-Exp (reasoning on)PromptingDeepSeek · best of 2 settings
85.5%
$0.27 / $0.41
33 Hammer2.1-7BFunction callingMadeAgents
85.5%
–
35 Grok 4 (0709)Function callingxAI · best of 2 settings
85.4%
–
36 Mistral Medium 3Mode not statedMistral AI · best of 2 settings
85.3%
$0.40 / $2
37 Falcon3 10B InstructFunction callingTII
85.0%
–
37 Hammer2.1-3BFunction callingMadeAgents
85.0%
–
39 CoALM 8BMode not statedUIUC and Oumi
84.9%
–
40 Mistral Large 2.1 (2411)Function callingMistral AI · best of 2 settings
84.7%
–
41 GPT-4.1 miniPromptingOpenAI · best of 2 settings
84.6%
$0.40 / $1.60
41 xLAM-2-8b-fc-rFunction callingSalesforce
84.6%
–
43 Llama 3.1 8B InstructPromptingMeta
84.0%
$0.05 / $0.08
44 CoALM 70BMode not statedUIUC and Oumi
83.4%
–
45 Hammer2.1-1.5BFunction callingMadeAgents
83.0%
–
45 xLAM-2-3b-fc-rFunction callingSalesforce
83.0%
–
47 Qwen3-1.7BFunction callingAlibaba
82.9%
–
48 Falcon3 7B InstructFunction callingTII
82.7%
–
48 Arch-Agent-1.5BMode not statedKatanemo
82.7%
–
48 Llama 3.2 3B InstructFunction callingMeta
82.7%
$0.05 / $0.33
51 Granite 20B FunctionCallingFunction callingIBM
82.4%
–
52 o3PromptingOpenAI · best of 2 settings
81.9%
$2 / $8
52 GPT-5.2Function callingOpenAI · best of 2 settings
81.9%
$1.75 / $14
54 MiniCPM3-4BFunction callingOpenBMB · best of 2 settings
81.8%
–
55 BitAgent Bounty 8BMode not statedBittensor
81.6%
–
55 Kimi K2 (0711)Function callingMoonshot AI
81.6%
$0.57 / $2.30
55 Nanbeige4-3B-Thinking (2511)Function callingNanbeige
81.6%
–
58 Bielik 11B v2.3 InstructPromptingSpeakLeash and ACK Cyfronet AGH
81.5%
–
59 o4-miniPromptingOpenAI · best of 2 settings
81.3%
–
60 Command R7B (12-2024)Function callingCohere
81.0%
$0.037 / $0.15
61 GPT-5 nanoPromptingOpenAI · best of 2 settings
80.8%
$0.05 / $0.40
62 Granite 3.2 8B InstructFunction callingIBM
79.8%
–
63 Gemma 3 12BPromptingGoogle
79.4%
$0.05 / $0.15
64 Granite 3.1 8B InstructFunction callingIBM
78.3%
–
65 Nova Micro 1.0Function callingAmazon
74.1%
$0.035 / $0.14
66 GPT-4.1 nanoFunction callingOpenAI · best of 2 settings
73.0%
$0.10 / $0.40
67 Qwen3-0.6BFunction callingAlibaba · best of 2 settings
71.8%
–
68 GPT-5 miniFunction callingOpenAI · best of 2 settings
69.9%
$0.25 / $2
69 Phi-4PromptingMicrosoft
69.6%
$0.07 / $0.14
70 xLAM-2-1b-fc-rFunction callingSalesforce
69.0%
–
71 RZN-TPromptingPhronetic AI
67.9%
–
71 Granite 4.0 350MFunction callingIBM
67.9%
–
73 Hammer2.1-0.5BFunction callingMadeAgents
66.0%
–
74 Gemma 3 4BPromptingGoogle
61.1%
$0.04 / $0.05
75 Falcon3 3B InstructFunction callingTII
54.6%
–
76 Llama 3.2 1B InstructFunction callingMeta
38.4%
$0.027 / $0.20
76 Nanbeige3.5-Pro-ThinkingFunction callingNanbeige
38.4%
–
78 Gemma 3 1BPromptingGoogle
20.2%
–
79 Falcon3 1B InstructFunction callingTII
9.0%
–
80 Ministral 8B (2410)Function callingMistral AI
0.0%
–
80 Llama 3.1 Nemotron Ultra 253B v1Function callingNVIDIA
0.0%
–

Swipe the table sideways for more columns.

Source: Berkeley Function Calling Leaderboard (BFCL) V4, via Spring Prompt, springprompt.com/benchmarks/bfcl. Test your own prompt on these models →

Ranks follow the score as shown, so equal numbers share a rank. Each model is shown at its best setting; show every setting. Results as published by Berkeley Function Calling Leaderboard (BFCL) V4; we do not re-run them.

Progress over time

Single-turn calls (curated), plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's; 38 models without a known release date are left off. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

0%50%100% Jul 20242025Jul2026Jul Mistral NemoQwen3-32BQwen3-235B-A22B-Instruct-2507Gemini 3 Pro Preview

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 7 others not drawn.

GoogleAlibabaMistral AIAnthropicMetaOpenAI

82%84%86%88%90%92% Jul 20242025Jul2026Jul GoogleAlibabaMistral AIAnthropicMetaOpenAI

What it measures

Choosing the right function and arguments in one turn, on BFCL's curated set.

What it does not measure

Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Berkeley Function Calling Leaderboard (BFCL) V4 by the UC Berkeley Gorilla team, Apache License 2.0.