Bench to the Future 3 — September Panel — Numeric
Forecasting
Numeric component of the BTF-3 September 2026 panel: 371 resolved forecasting questions evaluated by normalized ranked probability score against a frozen web corpus.
Top models (lower is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 0.1 |
| GPT-6 Astra | 0.1 |
| GPT-5.6 Sol | 0.1 |
| Opus 4.8 | 0.1 |
| Sonnet 5 | 0.1 |
| Muse Spark 1.3 | 0.1 |
| GLM-5.3 | 0.2 |
| GLM-5.3 Flash | 0.2 |