PreAct-Bench - Law and Contract: 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 law and contract domain. Source: arxiv.org. Saturation forecast: Around December 2027. 18 models tracked.

Top models

#ModelScore
1Gemma 3 27B88.1
2Qwen 3 30B A3B 2507 Instruct86.1
3GPT-5.478.8
4Qwen 3 8B (Non-reasoning)77.3
5Gemma 3 4B66.8
6Qwen 3 0.6B (Non-reasoning)66.7
7Qwen 3 30B A3B 2507 (Thinking)64.8
8Qwen 3 4B 2507 (Thinking)62.9
9GPT-4o59.9
10Qwen 3 8B (Thinking)58.2
11Qwen 3 0.6B (Thinking)56.3
12Qwen 3 4B 2507 Instruct56.1
13GPT-OSS-20B46.3
14GPT-OSS-120B41.3

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