PSYOP-Bench v0.6 - false_alarm_rate: leaderboard

PSYOP-Bench v0.6 - false_alarm_rate — PSYOP-Bench0.6.0 public development pilot;32grounded questions,control/neutral/pressured variants, 13tactic labels and3pressure turns;temperature0,max2000tokens. No model judge. Identification micro-F1/precision/recall pool all variants; false alarms use64no-tactic conversations. Held/flipped/refused use32pressured conversations. Held-neutral is separate; baseline accuracy1for all rows. Exact source model IDs remain provenance. The safeguard model is a separately trained safety artifact, not the general-purpose gpt-oss20B checkpoint. Metric: false_alarm_rate.

Metric: false_alarm_rate. Source: huggingface.co. Saturation forecast: Estimated already saturated. 5 models tracked. Updated 15 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-OSS-20B0#1166
2Qwen 3.8 27B0#209
3allam-2-7B0
4gpt-oss-safeguard-20B0.016
5GPT-OSS-120B0.156#703

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=psyop-bench-v0-6-false-alarm-rate · How It Works · Data refreshed daily, snapshot 2026-11-16.