Atlas

Benchmarks

← All benchmarks

Butter-Bench: Multi-Step Spatial Path Planning

Agents · 2025-10-23

Butter-Bench subtask measuring multi-step spatial path planning with routes broken into shorter navigation segments.

Top models (higher is better)

ModelScore
Human (Butter-Bench)100.0
Opus 4.160.0
Gemini 2.5 Pro40.0
Gemini Robotics-ER 1.520.0
GPT-50.0
Grok 40.0
Llama 4 Maverick0.0
Loading Atlas data…