SocialMaze - Social Graph Analysis: leaderboard
SocialMaze (Findings of EMNLP 2026, Table 4): given a social network's interaction traces, infer the relationship structure that explains them. 5-seed mean accuracy on 500 instances at temperature 0.7, higher is better. This is the one task with no human reference row in the paper's table.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 100 | #159 |
| 2 | O1 | 99.2 | #191 |
| 3 | O3 Mini | 99 | #283 |
| 4 | DeepSeek R1 | 98.6 | #265 |
| 5 | QwQ-32B | 95 | #422 |
| 6 | DeepSeek V3 | 92.4 | #333 |
| 7 | GPT-4o | 83.2 | #357 |
| 8 | Llama 3.3 70B | 81 | #578 |
| 9 | Qwen 2.5 72B | 80.6 | #514 |
| 10 | GPT-4o Mini | 53 | #599 |
| 11 | Phi-4 | 40.6 | #708 |
| 12 | Llama 3.1 8B | 28.2 | #1131 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=socialmaze-social-graph-analysis · How It Works · Data refreshed daily, snapshot 2026-11-06.