Bench to the Future 3 — Numeric
Forecasting · 2026-06-29
The numeric component of BTF-3 contains 392 resolved forecasting questions researched against a frozen web corpus. Distributional forecasts are evaluated by normalized ranked probability score (RPS), which generalizes the Brier score; lower scores are better.
Top models (lower is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 0.1 |
| GPT-5.5 | 0.1 |
| Claude Fable 5 | 0.1 |
| GPT-5.6 Sol | 0.1 |
| Opus 4.8 | 0.1 |
| Sonnet 5 | 0.2 |