SocialMemBench (Full Context, Small-Tier Subset): leaderboard
Memory questions about multi-party social group conversations, such as who said what, group decisions, member attributes, changes over time and exceptions to group norms; SocialMemBench (Owolabi, 2026). The model answers 77 questions about 10 small social networks with the whole conversation corpus in context; mean score in percent (multiple choice by exact letter match, open-ended answers by a GPT-4o-mini judge from 0 to 1), higher is better.
Metric: Mean score (%; question-weighted over 77 questions of 10 small-tier social networks, whole conversation corpus in context; multiple choice by exact letter match, open-ended by a GPT-4o-mini judge in 0-1). Source: arxiv.org. Saturation forecast: Around February 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 71.9 | #452 |
| 2 | GPT-4o Mini | 64.4 | #1627 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=socialmembench-full-context-small-tier-subset · How It Works · Data refreshed daily, snapshot 2026-11-12.