PreAct-Bench: 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; mean over the five domains (200 trajectories each). Source: arxiv.org. Saturation forecast: Around June 2028. 18 models tracked.

Top models

#ModelScore
1Gemma 3 27B83.9
2GPT-5.481.5
3Qwen 3 30B A3B 2507 Instruct79.6
4Qwen 3 8B (Non-reasoning)72.7
5Gemma 3 4B66.7
6Qwen 3 0.6B (Non-reasoning)66.7
7Qwen 3 8B (Thinking)62.7
8GPT-4o61.3
9Qwen 3 30B A3B 2507 (Thinking)58.8
10Qwen 3 0.6B (Thinking)58.5
11Qwen 3 4B 2507 Instruct51
12Qwen 3 4B 2507 (Thinking)49.3
13GPT-OSS-20B49.1
14GPT-OSS-120B45.5

Interactive version: theaggregate.ai/benchmark?slug=preact-bench · How It Works · Data refreshed daily, snapshot 2026-09-29.