Atlas

Benchmarks

VibePrevisBench Scene Check

VibePrevisBench's deterministic half: rule-based checks of each exported Blender scene against the task's hidden ground-truth facts, averaged over the 242 tasks, from the same runs as the overall score.

Models
9
Top score
70.0
Leader
GPT-6 Astra
Updated
Nov 11, 2026
Resolution
—
Saturation
—
Signal
80thpercentile
217 benchmarks from 0.8 to 3.7226 benchmarks from 3.7 to 6.6136 benchmarks from 6.6 to 9.5100 benchmarks from 9.5 to 12.451 benchmarks from 12.4 to 15.223 benchmarks from 15.2 to 18.115 benchmarks from 18.1 to 21.06 benchmarks from 21.0 to 23.97 benchmarks from 23.9 to 26.85 benchmarks from 26.8 to 29.73 benchmarks from 29.7 to 32.65 benchmarks from 32.6 to 35.43 benchmarks from 35.4 to 38.32 benchmarks from 38.3 to 41.21 benchmark from 41.2 to 44.11 benchmark from 47.0 to 49.91 benchmark from 49.9 to 52.81 benchmark from 67.2 to 70.1

163rd strongest of 803

Signal-to-noise +11.05 for the VibePrevisBench family

Difficulty
Not estimated

Another benchmark represents this family in the fit.

Loading Atlas data…