FutureSim: leaderboard
FutureSim final-day top-1 forecasting accuracy, separating quarters and agent harnesses. This card uses original Q1 Claude Code runs; Q2 has source-described cross-question leakage filtering. Handholding versions and all source seed metrics are preserved. Board: FutureSim. Q1 uses 330 questions; original leakage limitations are described by the source.
Metric: Final-day Top-1 Accuracy (%). Source: openforecaster.github.io. Saturation forecast: Rough model projection: around 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 (FutureSim Handholding: v1) | 26.52 | |
| 2 | Opus 4.7 (FutureSim Handholding: v3) | 21.21 | |
| 3 | Opus 4.6 (FutureSim Handholding: v1) | 18.79 | |
| 4 | Opus 4.8 (FutureSim Handholding: v1) | 18.18 | |
| 5 | GLM 5.1 (FutureSim Handholding: v1) | 12.52 | |
| 6 | DeepSeek V4 Pro (FutureSim Handholding: v3) | 11.92 | |
| 7 | GLM 5.2 (FutureSim Handholding: v3) | 8.69 |
Interactive version: theaggregate.ai/benchmark?slug=futuresim · How It Works · Data refreshed daily, snapshot 2026-10-20.