MAGI-Hard: leaderboard
Metric: Accuracy (%, 2024-05 snapshot). Source: huggingface.co. 59 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4o (2024-05-13) | 80.86 |
| 2 | GPT-4 (0613) | 77.85 |
| 3 | GPT-4 Turbo | 77.74 |
| 4 | GPT-4 Preview (0125) | 76.83 |
| 5 | Claude 3 Opus (20240229) | 76.55 |
| 6 | GPT-4 (0314) | 75.67 |
| 7 | GPT-4 Preview (1106) | 74.96 |
| 8 | Llama 3 70B Instruct | 67.97 |
| 9 | Mistral Large | 67.69 |
| 10 | Qwen 1.5 110B Chat | 66.09 |
| 11 | Yi 1.5 34B Chat | 64.85 |
| 12 | Llama 3 8B Instruct | 63.84 |
| 13 | Qwen 1.5 72B Chat | 63.47 |
| 14 | Mixtral 8x22B Instruct (v0.1) | 62.41 |
| 15 | Mistral Medium | 62.15 |
Interactive version: theaggregate.ai/benchmark?slug=magi-hard · How It Works · Data refreshed daily, snapshot 2026-09-05.