Benchmark RadarAI BENCHMARK PROFILE
SocietyBench
SocietyBench is a benchmark for forecasting counterfactual social-world evolution, collecting web news and social-media posts to build timelines and generating forecasting questions scored on probability calibration and temporal accuracy.
- Released
- 2026-08-04
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
Current benchmarks focus on task completion, not on social understanding and forecasting. SocietyBench measures how well models understand and predict social events, providing a complementary evaluation axis.
Motivation
Large language models (LLMs), and the agents built on top of them, are now benchmarked heavily on whether they can finish a task -- fix a bug, drive a browser, operate a GUI.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.