Bench to the Future 3 — Pooled Score
Forecasting · 2026-06-29
Bench to the Future 3 (BTF-3) is a pastcasting benchmark of 1,907 resolved forecasting questions researched against a frozen web corpus. Its pooled score combines 1,515 binary Brier scores and 392 numeric ranked probability scores, weighting each numeric question three times as much as each binary question; lower scores indicate better forecasts.
Top models (lower is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 0.1 |
| Opus 4.8 | 0.1 |
| Claude Fable 5 | 0.1 |
| GPT-5.5 | 0.1 |
| GPT-5.6 Sol | 0.1 |
| Sonnet 5 | 0.2 |