Atlas

Benchmarks

← All benchmarks

Bench to the Future 2 — Refinement

Forecasting · 2026-04-20

The refinement component of the BTF-2 Brier decomposition measures how decisively forecasts separate questions with different outcomes. Higher refinement is better.

Top models (higher is better)

ModelScore
Opus 4.60.1
Gemini 3.1 Pro Preview0.1
GPT-5.40.1
Grok 4.200.0
Loading Atlas data…