PreAct-Bench - Academic: 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 academic domain. Source: arxiv.org. Saturation forecast: Around December 2026. 18 models tracked.

Top models

#ModelScore
1Gemma 3 27B89.1
2Qwen 3 30B A3B 2507 Instruct88
3GPT-5.487.6
4Qwen 3 8B (Non-reasoning)84.9
5Qwen 3 8B (Thinking)77.7
6Qwen 3 30B A3B 2507 (Thinking)74.5
7GPT-OSS-20B71.4
8GPT-OSS-120B71
9GPT-4o70.6
10Qwen 3 4B 2507 Instruct67.1
11Gemma 3 4B66.8
12Qwen 3 0.6B (Non-reasoning)66.7
13Qwen 3 4B 2507 (Thinking)64.6
14Qwen 3 0.6B (Thinking)60.5

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