ImmersedPrivacy - Social Context Selection (Audio as Text): leaderboard

Metric: Selection accuracy (%; Tier 2: share of trials in which the model picks the candidate action with the highest mean human appropriateness rating for the observed social state; audio replaced by a text description of the soundscape (Tier 2) or the dialogue transcript (Tier 3)). Source: arxiv.org. Saturation forecast: Around 2029. 12 models tracked.

Top models

#ModelScore
1Qwen 3.5 27B (Thinking)65
2GPT-5.4 (2026-03-05)64
3Qwen 3.5 27B (Non-reasoning)56
4Gemini 3.1 Pro (Preview) (High)50
5Gemini 3 Flash (Preview) (High)50
6Gemini 3.1 Pro (Preview) (Low)50
7Gemini 3 Flash (Preview) (Low)47
8GPT-4o Mini (2024-07-18)39

Interactive version: theaggregate.ai/benchmark?slug=immersedprivacy-social-context-selection-audio-as-text · How It Works · Data refreshed daily, snapshot 2026-09-26.