File size: 3,317 Bytes
c5a66b5
040749c
 
c5a66b5
040749c
 
c5a66b5
 
 
 
 
23f7b04
 
ff57477
1dad967
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9feca0c
040749c
f5b3bc7
9feca0c
f5b3bc7
9feca0c
f5b3bc7
 
9feca0c
f5b3bc7
9feca0c
f5b3bc7
9feca0c
d20a508
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f5b3bc7
040749c
d2f8f7c
b3f02ae
9feca0c
f5b3bc7
 
 
d2f8f7c
 
 
 
 
 
 
 
 
040749c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
---
language:
- en
license: apache-2.0
size_categories:
- 1K<n<10K
task_categories:
- question-answering
tags:
- Reasoning
- LLM
- Encryption
- Decryption
configs:
- config_name: Rot13
  data_files:
  - split: test
    path: data/Rot13.jsonl
- config_name: Atbash
  data_files:
  - split: test
    path: data/Atbash.jsonl
- config_name: Polybius
  data_files:
  - split: test
    path: data/Polybius.jsonl
- config_name: Vigenere
  data_files:
  - split: test
    path: data/Vigenere.jsonl
- config_name: Reverse
  data_files:
  - split: test
    path: data/Reverse.jsonl
- config_name: SwapPairs
  data_files:
  - split: test
    path: data/SwapPairs.jsonl
- config_name: ParityShift
  data_files:
  - split: test
    path: data/ParityShift.jsonl
- config_name: DualAvgCode
  data_files:
  - split: test
    path: data/DualAvgCode.jsonl
- config_name: WordShift
  data_files:
  - split: test
    path: data/WordShift.jsonl
---

# CipherBank Benchmark

## Benchmark description

CipherBank, a comprehensive benchmark designed to evaluate the reasoning capabilities of LLMs in cryptographic decryption tasks. 
CipherBank comprises 2,358 meticulously crafted problems, covering 262 unique plaintexts across 5 domains and 14 subdomains, with a focus on privacy-sensitive and real-world scenarios that necessitate encryption. From a cryptographic perspective, CipherBank incorporates 3 major categories of encryption methods, spanning 9 distinct algorithms, ranging from classical ciphers to custom cryptographic techniques.

## Model Performance

We evaluate state-of-the-art LLMs on CipherBank, e.g., GPT-4o, DeepSeek-V3, and cutting-edge reasoning-focused models such as o1 and DeepSeek-R1. Our results reveal significant gaps in reasoning abilities not only between general-purpose chat LLMs and reasoning-focused LLMs but also in the performance of current reasoning-focused models when applied to classical cryptographic decryption tasks, highlighting the challenges these models face in understanding and manipulating encrypted data. 

| **Model** |  **CipherBank Score (%)**|
|--------------|----|
|Qwen2.5-72B-Instruct |0.55    |
|Llama-3.1-70B-Instruct  |0.38   |
|DeepSeek-V3 | 9.86   |
|GPT-4o-mini-2024-07-18 |   1.00 |
|GPT-4o-2024-08-06 | 8.82   |
|gemini-1.5-pro  | 9.54   |
|gemini-2.0-flash-exp   |  8.65|
|**Claude-Sonnet-3.5-1022**  |  **45.14**  |
|DeepSeek-R1  | 25.91   |
|gemini-2.0-flash-thinking  | 13.49   |
|o1-mini-2024-09-12  | 20.07   |
|**o1-2024-12-17** | **40.59**   |

## Please see paper & website for more information:
- [Paper](https://huggingface.co/papers/2504.19093)
- [https://arxiv.org/abs/2504.19093](https://arxiv.org/abs/2504.19093)
- [https://cipherbankeva.github.io/](https://cipherbankeva.github.io/)

## Citation
If you find CipherBank useful for your research and applications, please cite using this BibTeX:
```bibtex
@misc{li2025cipherbankexploringboundaryllm,
      title={CipherBank: Exploring the Boundary of LLM Reasoning Capabilities through Cryptography Challenges}, 
      author={Yu Li and Qizhi Pei and Mengyuan Sun and Honglin Lin and Chenlin Ming and Xin Gao and Jiang Wu and Conghui He and Lijun Wu},
      year={2025},
      eprint={2504.19093},
      archivePrefix={arXiv},
      primaryClass={cs.CR},
      url={https://arxiv.org/abs/2504.19093}, 
}
```