PreAct-Bench - Cybersecurity: leaderboard
Metric: Prefix Foresight F1 with the full prefix (PFF(100), binary F1 of the unethical class, 0-1 scaled to %): given an agent trajectory truncated just before its first overt unethical step, the monitor predicts whether the trajectory will turn unethical; paired ethical and unethical trajectories per task; default generation parameters; higher is better; the cybersecurity domain. Source: arxiv.org. Saturation forecast: Around November 2027. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemma 3 27B | 89 |
| 2 | GPT-5.4 | 85.5 |
| 3 | Qwen 3 30B A3B 2507 Instruct | 84.8 |
| 4 | Qwen 3 8B (Non-reasoning) | 76.2 |
| 5 | Gemma 3 4B | 66.7 |
| 6 | Qwen 3 0.6B (Non-reasoning) | 66.7 |
| 7 | Qwen 3 8B (Thinking) | 63.7 |
| 8 | Qwen 3 0.6B (Thinking) | 62.6 |
| 9 | Qwen 3 30B A3B 2507 (Thinking) | 60.7 |
| 10 | GPT-OSS-20B | 59.6 |
| 11 | GPT-4o | 51.9 |
| 12 | Qwen 3 4B 2507 Instruct | 51.9 |
| 13 | GPT-OSS-120B | 48.9 |
| 14 | Qwen 3 4B 2507 (Thinking) | 43 |
Interactive version: theaggregate.ai/benchmark?slug=preact-bench-cybersecurity · How It Works · Data refreshed daily, snapshot 2026-09-29.