KOR-Bench - Cipher: leaderboard

Metric: Accuracy (%; zero-shot, 250 questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 38 models tracked.

Top models

#ModelScore
1O1 Preview (2024-09-12)82.8
2O1 Mini (2024-09-12)79.6
3GPT-4o (2024-05-13)42.8
4Claude 3.5 Sonnet (20240620)33.2
5Llama 3.1 405B Instruct31.2
6Qwen 2.5 32B Instruct26.8
7Qwen 2.5 72B Instruct26.4
8GPT-4 Turbo23.2
9Mistral Large 2 (Jul)22.8
10Yi Large20.8
11Llama 3.1 70B Instruct20.4
12Llama 3 70B Instruct20.4
13DeepSeek V2.518
14Qwen 2.5 14B Instruct14.4
15Qwen 2 72B Instruct12.8

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