CipherBank: leaderboard

Metric: Decryption Accuracy (%; 3-shot, mean of nine ciphers). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.

Top models

#ModelScore
1Claude 3.5 Sonnet (20241022)45.14
2O1 (2024-12-17)40.59
3DeepSeek R125.91
4O1 Mini (2024-09-12)20.07
5Gemini 2.0 Flash (Thinking)13.49
6DeepSeek V39.86
7Gemini 1.5 Pro9.54
8GPT-4o (2024-08-06)8.82
9Gemini 2.0 Flash (Preview)8.65
10GPT-4o (2024-11-20)6.4
11GPT-4o Mini (2024-07-18)1
12QwQ 32B-Preview0.76
13Qwen 2.5 72B Instruct0.55
14Llama 3.3 70B Instruct0.42
15Llama 3.1 70B Instruct0.38

Interactive version: theaggregate.ai/benchmark?slug=cipherbank · How It Works · Data refreshed daily, snapshot 2026-09-24.