TriggerBench - Positive Overloaded PM Accuracy: leaderboard

Metric: PM accuracy (%) on the 488 positive tasks whose trigger turn also carries an unrelated request; base dialogues of about 2.5K tokens; TriggerBench: 1,265 prospective-memory tasks from 488 dialogue blueprints in five dimensions (state tracking, temporal grounding, logical adherence, attention recovery, safe coding); a constraint is stated early in a dialogue and a later trigger turn calls for acting on it without being asked; GPT-4o (T=0) judges whether the reply fulfils the proactive intent; temperature 0.6; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 9 models tracked.

Top models

#ModelScore
1GPT-5.2 (High)91.6
2GPT-5.2 (Medium)87.7
3Gemma 3 27B (IT)68.85
4GPT-5.2 (Non-reasoning)64.55
5Qwen 3 235B A22B 2507 FP8 (Thinking)64.14
6GPT-4.136.68
7Qwen 3 235B A22B 2507 Instruct35.66
8GPT-4o22.34
9Qwen 3 32B13.11

Interactive version: theaggregate.ai/benchmark?slug=triggerbench-positive-overloaded-pm-accuracy · How It Works · Data refreshed daily, snapshot 2026-09-29.