Benchmarks / Berkeley Function Calling Leaderboard (BFCL) V4

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

Berkeley Function Calling Leaderboard (BFCL) V4

Storing and recalling facts across sessions with memory tools.

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: Storing and recalling facts across sessions with memory tools.

Latest results. , 81 models ranked.

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

  1. Claude Opus 4.5 (Anthropic): 73.8%
  2. Claude Sonnet 4.5 (Anthropic): 65.0%
  3. Gemini 3 Pro Preview (Google): 61.7%

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.

Memory: Qwen3-1.7B

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

  1. 1 Claude Opus 4.5Function callingAnthropic 73.8%
  2. 2 Claude Sonnet 4.5Function callingAnthropic 65.0%
  3. 3 Gemini 3 Pro PreviewPromptingGoogle 61.7%
  4. 4 Grok 4 (0709)Function callingxAI 55.9%
  5. 5 GLM-4.6 (reasoning on)Function callingZ.ai 55.7%
  6. 6 Claude Haiku 4.5Function callingAnthropic 54.4%
  7. 7 DeepSeek-V3.2-ExpFunction callingDeepSeek 54.2%
  8. 8 Grok 4.1 Fast (reasoning)Function callingxAI 54.0%
  9. 9 o3PromptingOpenAI 51.8%
  10. 10 GPT-5.2Function callingOpenAI 45.8%
  11. 11 Nanbeige3.5-Pro-ThinkingFunction callingNanbeige 45.2%
  12. 12 GPT-5 miniFunction callingOpenAI 44.3%
  13. 13 Gemini 2.5 FlashFunction callingGoogle 41.3%
  14. 14 Nanbeige4-3B-Thinking (2511)Function callingNanbeige 36.8%
  15. 15 o4-miniPromptingOpenAI 35.3%
  16. 62 Qwen3-1.7BFunction callingAlibaba 6.0%

Full results

BFCL: memory, % correct, higher is better
#ModelMemory
% correct, higher is better
Price
$ per million tokens, in / out
1 Claude Opus 4.5Function callingAnthropic · best of 2 settings
73.8%
$5 / $25
2 Claude Sonnet 4.5Function callingAnthropic · best of 2 settings
65.0%
$3 / $15
3 Gemini 3 Pro PreviewPromptingGoogle · best of 2 settings
61.7%
–
4 Grok 4 (0709)Function callingxAI · best of 2 settings
55.9%
–
5 GLM-4.6 (reasoning on)Function callingZ.ai
55.7%
$0.50 / $2
6 Claude Haiku 4.5Function callingAnthropic · best of 2 settings
54.4%
$1 / $5
7 DeepSeek-V3.2-ExpFunction callingDeepSeek · best of 2 settings
54.2%
$0.27 / $0.41
8 Grok 4.1 Fast (reasoning)Function callingxAI
54.0%
–
9 o3PromptingOpenAI · best of 2 settings
51.8%
$2 / $8
10 GPT-5.2Function callingOpenAI · best of 2 settings
45.8%
$1.75 / $14
11 Nanbeige3.5-Pro-ThinkingFunction callingNanbeige
45.2%
–
12 GPT-5 miniFunction callingOpenAI · best of 2 settings
44.3%
$0.25 / $2
13 Gemini 2.5 FlashFunction callingGoogle · best of 2 settings
41.3%
$0.30 / $2.50
14 Nanbeige4-3B-Thinking (2511)Function callingNanbeige
36.8%
–
15 o4-miniPromptingOpenAI · best of 2 settings
35.3%
–
16 Kimi K2 (0711)Function callingMoonshot AI
29.0%
$0.57 / $2.30
17 Command A ReasoningFunction callingCohere
28.8%
–
18 Gemma 3 12BPromptingGoogle
27.5%
$0.05 / $0.15
18 Falcon3 10B InstructFunction callingTII
27.5%
–
20 GPT-4.1 miniFunction callingOpenAI · best of 2 settings
26.9%
$0.40 / $1.60
21 Qwen3-32BFunction callingAlibaba · best of 2 settings
26.7%
$0.14 / $0.40
22 Grok 4.1 Fast (non-reasoning)Function callingxAI
26.2%
–
23 Mistral Large 2.1 (2411)Function callingMistral AI · best of 2 settings
25.0%
–
24 Phi-4PromptingMicrosoft
24.7%
$0.07 / $0.14
24 GPT-5 nanoFunction callingOpenAI · best of 2 settings
24.7%
$0.05 / $0.40
26 Qwen3-235B-A22B-Instruct-2507Function callingAlibaba · best of 2 settings
23.9%
$0.15 / $0.75
26 Qwen3-4B-Instruct-2507PromptingAlibaba · best of 2 settings
23.9%
–
26 GPT-4.1Function callingOpenAI · best of 2 settings
23.9%
$2 / $8
29 Mistral Medium 3Function callingMistral AI · best of 2 settings
23.0%
$0.40 / $2
30 xLAM-2-32b-fc-rFunction callingSalesforce
20.9%
–
31 Gemini 2.5 Flash-LiteFunction callingGoogle · best of 2 settings
20.7%
$0.10 / $0.40
31 Falcon3 7B InstructFunction callingTII
20.7%
–
33 Qwen3-14BFunction callingAlibaba · best of 2 settings
19.6%
$0.10 / $0.22
34 Llama 4 Maverick 17B 128E Instruct (FP8)Function callingMeta
18.9%
–
34 GPT-4.1 nanoFunction callingOpenAI · best of 2 settings
18.9%
$0.10 / $0.40
36 ToolACE-2-8BFunction callingHuawei Noah's Ark Lab and USTC
18.5%
–
37 Mistral Small 3.2 24BFunction callingMistral AI · best of 2 settings
18.1%
$0.094 / $0.25
38 Qwen3-30B-A3B-Instruct-2507Function callingAlibaba · best of 2 settings
17.6%
$0.09 / $0.30
39 Command AFunction callingCohere
16.6%
$2.50 / $10
40 Qwen3-8BFunction callingAlibaba · best of 2 settings
14.6%
–
40 Arch-Agent-32BMode not statedKatanemo
14.6%
–
42 Granite 3.1 8B InstructFunction callingIBM
14.4%
–
42 xLAM-2-70b-fc-rFunction callingSalesforce
14.4%
–
44 xLAM-2-8b-fc-rFunction callingSalesforce
14.0%
–
45 Gemma 3 27BPromptingGoogle
13.6%
$0.08 / $0.35
46 Palmyra X 004Function callingWriter
13.1%
–
47 Granite 3.2 8B InstructFunction callingIBM
12.5%
–
48 MiniCPM3-4BFunction callingOpenBMB · best of 2 settings
12.0%
–
49 xLAM-2-3b-fc-rFunction callingSalesforce
11.4%
–
49 Bielik 11B v2.3 InstructPromptingSpeakLeash and ACK Cyfronet AGH
11.4%
–
51 Llama 3.1 8B InstructPromptingMeta
10.8%
$0.05 / $0.08
52 Mistral NemoFunction callingMistral AI · best of 2 settings
10.3%
$0.029 / $0.03
53 Qwen3-0.6BFunction callingAlibaba · best of 2 settings
8.6%
–
53 Gemma 3 4BPromptingGoogle
8.6%
$0.04 / $0.05
55 Arch-Agent-1.5BMode not statedKatanemo
8.2%
–
55 Llama 3.3 70B InstructFunction callingMeta
8.2%
$0.71 / $0.71
55 Llama 4 ScoutFunction callingMeta
8.2%
$0.10 / $0.30
58 Falcon3 3B InstructFunction callingTII
7.7%
–
59 Arch-Agent-3BMode not statedKatanemo
6.9%
–
59 RZN-TPromptingPhronetic AI
6.9%
–
61 Llama 3.2 3B InstructFunction callingMeta
6.2%
$0.05 / $0.33
62 Qwen3-1.7BFunction callingAlibaba
6.0%
–
63 Falcon3 1B InstructFunction callingTII
5.8%
–
63 CoALM 70BMode not statedUIUC and Oumi
5.8%
–
65 Command R7B (12-2024)Function callingCohere
5.2%
$0.037 / $0.15
66 Ministral 8B (2410)Function callingMistral AI
4.5%
–
67 xLAM-2-1b-fc-rFunction callingSalesforce
3.9%
–
68 Gemma 3 1BPromptingGoogle
3.2%
–
68 Granite 4.0 350MFunction callingIBM
3.2%
–
68 Llama 3.2 1B InstructFunction callingMeta
3.2%
$0.027 / $0.20
71 Hammer2.1-3BFunction callingMadeAgents
3.0%
–
72 CoALM 8BMode not statedUIUC and Oumi
2.8%
–
73 Nova 2 LiteFunction callingAmazon
2.4%
$0.30 / $2.50
73 Nova Micro 1.0Function callingAmazon
2.4%
$0.035 / $0.14
75 Nova Pro 1.0Function callingAmazon
1.9%
$0.80 / $3.20
76 BitAgent Bounty 8BMode not statedBittensor
1.5%
–
77 Hammer2.1-0.5BFunction callingMadeAgents
1.1%
–
78 Granite 20B FunctionCallingFunction callingIBM
0.0%
–
78 Hammer2.1-1.5BFunction callingMadeAgents
0.0%
–
78 Hammer2.1-7BFunction callingMadeAgents
0.0%
–
78 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

Memory, 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%0%20%40%60%80% Jul 20242025Jul2026Jul Mistral NemoLlama 3.1 8B InstructMistral Large 2.1 (2411)Gemma 3 12Bo3Claude Sonnet 4.5Claude Opus 4.5

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.

AnthropicGoogleZ.aiDeepSeekxAIOpenAI

20%40%60%80% Jul 20242025Jul2026Jul AnthropicGoogleZ.aiDeepSeekxAIOpenAI

What it measures

Storing and recalling facts across sessions with memory tools.

What it does not measure

Not long-term personal memory in a product: sessions are BFCL's scripted ones.

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