SocialPersona - Extractive-Abstractive Profiling: leaderboard
SocialPersona (2026) tests whether multimodal models can infer what a person cares about from a real social-media timeline (post text, image captions and timestamps) and use it. This board is interest-tag F1 in percent when the model first selects representative posts per domain and then writes the profile from them alone; higher is better.
Metric: Interest-tag F1 (%) of the constructed profile against human-verified stable and recent interest tags in seven domains (exact match, then o3 semantic matching), from timelines of 100 social-media users given as post text, image captions and timestamps (up to 200 posts), the model first selects up to 12 representative posts per domain and then writes the profile from them alone; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 37.48 | #123 |
| 2 | Qwen 3.5 35B A3B | 26.77 | #482 |
| 3 | GPT-4o Mini | 21.59 | #1623 |
| 4 | Qwen 3 VL 8B Instruct | 8.34 | #981 |
| 5 | Gemini 2.5 Flash | 6.33 | #453 |
| 6 | Qwen 2.5 VL 7B Instruct | 3.09 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=socialpersona-extractive-abstractive-profiling · How It Works · Data refreshed daily, snapshot 2026-11-13.