KOR-Bench - Puzzle: leaderboard
KOR-Bench puzzles solved under 25 newly written puzzle rules. Chat models, zero-shot. Higher is better. Source: arxiv.org/abs/2410.06526.
Metric: Accuracy (%; zero-shot, 250 questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 Preview (2024-09-12) | 36.8 | #220 |
| 2 | O1 Mini (2024-09-12) | 35.6 | #414 |
| 3 | GPT-4o (2024-05-13) | 16.8 | #338 |
| 4 | Mistral Large 2 (Jul) | 15.6 | #498 |
| 5 | Claude 3.5 Sonnet (20240620) | 14.8 | #309 |
| 6 | Llama 3.1 405B Instruct | 13.93 | #472 |
| 7 | GPT-4 Turbo | 12.8 | #419 |
| 8 | Yi Large | 11.2 | #565 |
| 9 | DeepSeek V2.5 | 11.2 | #503 |
| 10 | Gemini 1.5 Pro | 10.8 | #453 |
| 11 | Qwen 2.5 72B Instruct | 10.4 | #470 |
| 12 | Qwen 2 72B Instruct | 9.2 | #584 |
| 13 | Qwen 2.5 14B Instruct | 9.2 | #657 |
| 14 | Qwen 2.5 32B Instruct | 8 | #511 |
| 15 | Llama 3.1 70B Instruct | 7.6 | #570 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kor-bench-puzzle · How It Works · Data refreshed daily, snapshot 2026-11-07.