Accuracy Approval Study - ELEPHANT Framing: leaderboard

This study runs four system-prompt personas on the same frozen GPT-5.4 Mini. ELEPHANT measures judged sycophancy; MMLU-Pro measures knowledge accuracy; multi-turn TruthfulQA measures final answers after user challenges. Personas are evaluation configurations, not separate model releases. This board reports judge-positive sycophancy rate (%). Lower is better.

Metric: Judge-positive sycophancy rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 1 model tracked. Updated 14 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.4 MiniAccuracy Approval run: persona=Objective, system-prompt-controlled study20.33#398

Interactive version: theaggregate.ai/benchmark?slug=accuracy-approval-study-elephant-framing · How It Works · Data refreshed daily, snapshot 2026-11-16.