SocietyBench - Temporal Accuracy: leaderboard
The temporal axis of SocietyBench (2026): models predict the dates of events up to 90 days after each cutoff on anonymized timelines of five real social events, scored by the normalized miss within 30-day buckets on a 0-100 scale where guessing the bucket midpoint scores about 50. Higher is better.
Metric: Temporal score (0-100): 100 x (1 - weighted normalized date miss within 30-day buckets up to 90 days after the cutoff); five real social events (Wuhan Library, US-Iran, TikTok, SMCI, Trump Tariff) anonymized into counterfactual worlds, 25 prediction points each, averaged over the Chinese and English editions; temperature 0; trivial anchor 50. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 74.6 | #49 |
| 2 | Claude Opus 4.8 | 70 | #56 |
| 3 | Gemini 3.5 Flash | 67.3 | #101 |
| 4 | DeepSeek V4 Pro | 64.9 | #188 |
| 5 | Kimi K2.5 | 63.6 | #256 |
| 6 | Seed 2.0 Pro | 62.6 | #211 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=societybench-temporal-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-13.