PASTABench: leaderboard

Metric: Perfect interruption rate (%; share of the 1,139 risky multi-turn agent trajectories on which the monitor first interrupts inside the optimal intervention window, at or after the earliest-signal turn and before the trigger turn, and names the correct level-2 risk sub-category; zero-shot, greedy decoding, stepwise auto-stop protocol). Source: arxiv.org. Saturation forecast: Around 2032. 16 models tracked.

Top models

#ModelScore
1GPT-4o40.74
2Gemini 3 Pro39.42
3Llama 4 Maverick33.36
4Llama 4 Scout30.99
5DeepSeek R130.47
6Grok 329.94
7Llama 3.3 70B Instruct24.06
8Gemini 2.5 Pro23.79
9QwQ-32B22.65

Interactive version: theaggregate.ai/benchmark?slug=pastabench · How It Works · Data refreshed daily, snapshot 2026-09-26.