Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

Berkeley Function Calling Leaderboard (BFCL) V4

Declining to call a function when none of those offered fits.

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: Declining to call a function when none of those offered fits.

Latest results. , 81 models ranked.

Top 3 on Irrelevance detection (% correct, higher is better, each model at its best setting):

  1. Ministral 8B (2410) (Mistral AI): 100.0%
  2. Llama 3.1 Nemotron Ultra 253B v1 (NVIDIA): 100.0%
  3. BitAgent Bounty 8B (Bittensor): 97.5%

The first 2 are too close to tell apart, so read the order among them as a tie.

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.

Irrelevance detection: Llama 4 Scout

Top 15 of 81 results · % correct, higher is better · ≈ cannot be told apart from the leader. Choose a model to highlight it.Clear highlight

  1. 1≈ Ministral 8B (2410)Function callingMistral AI 100.0%
  2. 1≈ Llama 3.1 Nemotron Ultra 253B v1Function callingNVIDIA 100.0%
  3. 3 BitAgent Bounty 8BMode not statedBittensor 97.5%
  4. 4 Claude Haiku 4.5PromptingAnthropic 95.3%
  5. 5 Claude Sonnet 4.5PromptingAnthropic 95.0%
  6. 6 Gemini 2.5 FlashFunction callingGoogle 93.7%
  7. 7 Gemini 2.5 Flash-LitePromptingGoogle 93.3%
  8. 8 DeepSeek-V3.2-ExpFunction callingDeepSeek 93.2%
  9. 9 Mistral Medium 3Function callingMistral AI 92.0%
  10. 10 GPT-5 miniFunction callingOpenAI 91.0%
  11. 11 ToolACE-2-8BFunction callingHuawei Noah's Ark Lab and USTC 90.8%
  12. 11 Claude Opus 4.5PromptingAnthropic 90.8%
  13. 13 Hammer2.1-7BFunction callingMadeAgents 90.1%
  14. 14 GPT-5 nanoFunction callingOpenAI 89.1%
  15. 15 Mistral Small 3.2 24BFunction callingMistral AI 87.9%
  16. 75 Llama 4 ScoutFunction callingMeta 44.9%

Full results

BFCL: irrelevance detection, % correct, higher is better
#ModelIrrelevance detection
% correct, higher is better
Price
$ per million tokens, in / out
1≈ Ministral 8B (2410)Function callingMistral AI
100.0%
–
1≈ Llama 3.1 Nemotron Ultra 253B v1Function callingNVIDIA
100.0%
–
3 BitAgent Bounty 8BMode not statedBittensor
97.5%
–
4 Claude Haiku 4.5PromptingAnthropic · best of 2 settings
95.3%
$1 / $5
5 Claude Sonnet 4.5PromptingAnthropic · best of 2 settings
95.0%
$3 / $15
6 Gemini 2.5 FlashFunction callingGoogle · best of 2 settings
93.7%
$0.30 / $2.50
7 Gemini 2.5 Flash-LitePromptingGoogle · best of 2 settings
93.3%
$0.10 / $0.40
8 DeepSeek-V3.2-ExpFunction callingDeepSeek · best of 2 settings
93.2%
$0.27 / $0.41
9 Mistral Medium 3Function callingMistral AI · best of 2 settings
92.0%
$0.40 / $2
10 GPT-5 miniFunction callingOpenAI · best of 2 settings
91.0%
$0.25 / $2
11 ToolACE-2-8BFunction callingHuawei Noah's Ark Lab and USTC
90.8%
–
11 Claude Opus 4.5PromptingAnthropic · best of 2 settings
90.8%
$5 / $25
13 Hammer2.1-7BFunction callingMadeAgents
90.1%
–
14 GPT-5 nanoFunction callingOpenAI · best of 2 settings
89.1%
$0.05 / $0.40
15 Mistral Small 3.2 24BFunction callingMistral AI · best of 2 settings
87.9%
$0.094 / $0.25
16 Phi-4PromptingMicrosoft
87.6%
$0.07 / $0.14
17 Kimi K2 (0711)Function callingMoonshot AI
87.3%
$0.57 / $2.30
17 Falcon3 1B InstructFunction callingTII
87.3%
–
17 GPT-5.2PromptingOpenAI · best of 2 settings
87.3%
$1.75 / $14
20 Qwen3-14BPromptingAlibaba · best of 2 settings
87.2%
$0.10 / $0.22
20 o4-miniPromptingOpenAI · best of 2 settings
87.2%
–
22 CoALM 8BMode not statedUIUC and Oumi
86.9%
–
23 Command A ReasoningFunction callingCohere
86.8%
–
24 GPT-4.1Function callingOpenAI · best of 2 settings
86.5%
$2 / $8
25 o3Function callingOpenAI · best of 2 settings
86.1%
$2 / $8
25 Hammer2.1-3BFunction callingMadeAgents
86.1%
–
27 CoALM 70BMode not statedUIUC and Oumi
85.7%
–
28 Gemini 3 Pro PreviewPromptingGoogle · best of 2 settings
85.6%
–
29 GLM-4.6 (reasoning on)Function callingZ.ai
85.0%
$0.50 / $2
30 Qwen3-4B-Instruct-2507Function callingAlibaba · best of 2 settings
84.9%
–
31 Grok 4 (0709)PromptingxAI · best of 2 settings
84.3%
–
32 Command AFunction callingCohere
84.2%
$2.50 / $10
33 GPT-4.1 nanoPromptingOpenAI · best of 2 settings
83.4%
$0.10 / $0.40
34 Nanbeige4-3B-Thinking (2511)Function callingNanbeige
83.1%
–
35 Qwen3-0.6BPromptingAlibaba · best of 2 settings
82.5%
–
36 RZN-TPromptingPhronetic AI
82.4%
–
36 Qwen3-32BPromptingAlibaba · best of 2 settings
82.4%
$0.14 / $0.40
38 Qwen3-8BPromptingAlibaba · best of 2 settings
82.3%
–
39 Arch-Agent-32BMode not statedKatanemo
82.2%
–
40 Nova 2 LiteFunction callingAmazon
82.1%
$0.30 / $2.50
41 Qwen3-235B-A22B-Instruct-2507Function callingAlibaba · best of 2 settings
81.7%
$0.15 / $0.75
41 GPT-4.1 miniFunction callingOpenAI · best of 2 settings
81.7%
$0.40 / $1.60
41 Command R7B (12-2024)Function callingCohere
81.7%
$0.037 / $0.15
44 Palmyra X 004Function callingWriter
81.0%
–
45 Hammer2.1-0.5BFunction callingMadeAgents
80.8%
–
46 Granite 3.2 8B InstructFunction callingIBM
80.5%
–
47 xLAM-2-32b-fc-rFunction callingSalesforce
80.2%
–
48 Granite 3.1 8B InstructFunction callingIBM
80.0%
–
49 Qwen3-30B-A3B-Instruct-2507Function callingAlibaba · best of 2 settings
79.9%
$0.09 / $0.30
50 Grok 4.1 Fast (reasoning)Function callingxAI
79.4%
–
50 Hammer2.1-1.5BFunction callingMadeAgents
79.4%
–
52 xLAM-2-70b-fc-rFunction callingSalesforce
79.1%
–
53 Qwen3-1.7BFunction callingAlibaba
76.5%
–
54 Granite 20B FunctionCallingFunction callingIBM
75.1%
–
55 Arch-Agent-1.5BMode not statedKatanemo
74.8%
–
56 Arch-Agent-3BMode not statedKatanemo
74.7%
–
57 Nanbeige3.5-Pro-ThinkingFunction callingNanbeige
74.2%
–
58 Grok 4.1 Fast (non-reasoning)Function callingxAI
74.1%
–
59 MiniCPM3-4BPromptingOpenBMB · best of 2 settings
73.7%
–
59 Gemma 3 27BPromptingGoogle
73.7%
$0.08 / $0.35
61 Nova Micro 1.0Function callingAmazon
70.7%
$0.035 / $0.14
62 Gemma 3 12BPromptingGoogle
70.3%
$0.05 / $0.15
63 Nova Pro 1.0Function callingAmazon
70.1%
$0.80 / $3.20
64 Mistral Large 2.1 (2411)Function callingMistral AI · best of 2 settings
68.9%
–
65 xLAM-2-1b-fc-rFunction callingSalesforce
64.5%
–
66 xLAM-2-3b-fc-rFunction callingSalesforce
63.5%
–
67 xLAM-2-8b-fc-rFunction callingSalesforce
63.3%
–
68 Mistral NemoFunction callingMistral AI · best of 2 settings
61.8%
$0.029 / $0.03
69 Granite 4.0 350MFunction callingIBM
60.8%
–
70 Llama 4 Maverick 17B 128E Instruct (FP8)Function callingMeta
56.0%
–
71 Gemma 3 4BPromptingGoogle
53.9%
$0.04 / $0.05
72 Llama 3.3 70B InstructFunction callingMeta
53.5%
$0.71 / $0.71
73 Llama 3.2 3B InstructFunction callingMeta
52.1%
$0.05 / $0.33
74 Llama 3.2 1B InstructFunction callingMeta
51.6%
$0.027 / $0.20
75 Llama 4 ScoutFunction callingMeta
44.9%
$0.10 / $0.30
76 Llama 3.1 8B InstructPromptingMeta
42.7%
$0.05 / $0.08
77 Bielik 11B v2.3 InstructPromptingSpeakLeash and ACK Cyfronet AGH
36.0%
–
78 Gemma 3 1BPromptingGoogle
33.2%
–
79 Falcon3 3B InstructFunction callingTII
32.9%
–
80 Falcon3 10B InstructFunction callingTII
32.1%
–
81 Falcon3 7B InstructFunction callingTII
32.0%
–

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

Irrelevance detection, 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.

20%40%60%80%100%120% Jul 20242025Jul2026Jul Mistral NemoMinistral 8B (2410)

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.

Mistral AIAnthropicGoogleDeepSeekOpenAIMicrosoft

60%80%100% Jul 20242025Jul2026Jul Mistral AIAnthropicGoogleDeepSeekOpenAIMicrosoft

What it measures

Declining to call a function when none of those offered fits.

What it does not measure

Not general refusal or safety behaviour.

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