KOR-Bench - Operation: leaderboard
KOR-Bench questions under 25 newly defined mathematical operators and symbols, applied exactly as the rule states. 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 | Qwen 2.5 32B Instruct | 93.2 | #511 |
| 2 | GPT-4 Turbo | 90.4 | #419 |
| 3 | O1 Preview (2024-09-12) | 88.8 | #220 |
| 4 | Claude 3.5 Sonnet (20240620) | 88.4 | #309 |
| 5 | Llama 3.1 405B Instruct | 87.82 | #472 |
| 6 | Mistral Large 2 (Jul) | 86.8 | #498 |
| 7 | GPT-4o (2024-05-13) | 86 | #338 |
| 8 | Llama 3.1 70B Instruct | 84.8 | #570 |
| 9 | Qwen 2.5 14B Instruct | 84.4 | #657 |
| 10 | Yi Large | 84 | #565 |
| 11 | Qwen 2.5 72B Instruct | 83.6 | #470 |
| 12 | O1 Mini (2024-09-12) | 82.8 | #414 |
| 13 | Llama 3 70B Instruct | 82.4 | #643 |
| 14 | Gemini 1.5 Pro | 81.6 | #453 |
| 15 | Yi 1.5 34B Chat | 79.6 | #802 |
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-operation · How It Works · Data refreshed daily, snapshot 2026-11-07.