RedCode - Exec Python (ReAct): leaderboard
Metric: Attack success rate (%). Source: redcode-agent.github.io. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3 8B Instruct | 42.5 |
| 2 | Llama 3.1 8B Instruct | 62.87 |
| 3 | GPT-4 Preview (1106) | 64.5 |
| 4 | Claude 3.5 Sonnet (20240620) | 67.63 |
| 5 | GPT-3.5 Turbo (1106) | 76.23 |
| 6 | Llama 3.1 70B Instruct | 76.7 |
| 7 | GPT-4o (2024-05-13) | 77.23 |
| 8 | CodeQwen1.5-7B Chat | 77.57 |
| 9 | deepseek-coder-6.7B-instruct | 80.23 |
Interactive version: theaggregate.ai/benchmark?slug=redcode-exec-python-react · How It Works · Data refreshed daily, snapshot 2026-10-09.